Ce cahier vous apprend à entraîner un modèle de classification de pose à l'aide de MoveNet et de TensorFlow Lite. Le résultat est un nouveau modèle TensorFlow Lite qui accepte la sortie du modèle MoveNet comme entrée et génère une classification de pose, telle que le nom d'une pose de yoga.
La procédure dans ce cahier se compose de 3 parties :
- Partie 1 : Prétraitez les données d'apprentissage de la classification des poses dans un fichier CSV qui spécifie les points de repère (points clés du corps) détectés par le modèle MoveNet, ainsi que les étiquettes de pose de vérité terrain.
- Partie 2 : créez et entraînez un modèle de classification de pose qui prend les coordonnées du point de repère du fichier CSV en entrée et génère les étiquettes prédites.
- Partie 3 : Convertissez le modèle de classification de pose en TFLite.
Par défaut, ce bloc-notes utilise un ensemble de données d'images avec des poses de yoga étiquetées, mais nous avons également inclus une section dans la partie 1 où vous pouvez télécharger votre propre ensemble de données d'images de poses.
 Voir sur TensorFlow.org |  Exécuter dans Google Colab |  Voir la source sur GitHub |  Télécharger le cahier |  Voir le modèle TF Hub |
Préparation
Dans cette section, vous importerez les bibliothèques nécessaires et définirez plusieurs fonctions pour prétraiter les images d'entraînement dans un fichier CSV contenant les coordonnées des points de repère et les étiquettes de vérité terrain.
Rien d'observable ne se produit ici, mais vous pouvez développer les cellules de code cachées pour voir l'implémentation de certaines des fonctions que nous appellerons plus tard.
Si vous souhaitez uniquement créer le fichier CSV sans connaître tous les détails, exécutez simplement cette section et passez à la partie 1.
pip install -q opencv-python
import csv
import cv2
import itertools
import numpy as np
import pandas as pd
import os
import sys
import tempfile
import tqdm
from matplotlib import pyplot as plt
from matplotlib.collections import LineCollection
import tensorflow as tf
import tensorflow_hub as hub
from tensorflow import keras
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, classification_report, confusion_matrix
Code pour exécuter l'estimation de pose à l'aide de MoveNet
Fonctions pour exécuter l'estimation de pose avec MoveNet
# Download model from TF Hub and check out inference code from GitHub
!wget -q -O movenet_thunder.tflite https://tfhub.dev/google/lite-model/movenet/singlepose/thunder/tflite/float16/4?lite-format=tflite
!git clone https://github.com/tensorflow/examples.git
pose_sample_rpi_path = os.path.join(os.getcwd(), 'examples/lite/examples/pose_estimation/raspberry_pi')
sys.path.append(pose_sample_rpi_path)
# Load MoveNet Thunder model
import utils
from data import BodyPart
from ml import Movenet
movenet = Movenet('movenet_thunder')
# Define function to run pose estimation using MoveNet Thunder.
# You'll apply MoveNet's cropping algorithm and run inference multiple times on
# the input image to improve pose estimation accuracy.
def detect(input_tensor, inference_count=3):
"""Runs detection on an input image.
Args:
input_tensor: A [height, width, 3] Tensor of type tf.float32.
Note that height and width can be anything since the image will be
immediately resized according to the needs of the model within this
function.
inference_count: Number of times the model should run repeatly on the
same input image to improve detection accuracy.
Returns:
A Person entity detected by the MoveNet.SinglePose.
"""
image_height, image_width, channel = input_tensor.shape
# Detect pose using the full input image
movenet.detect(input_tensor.numpy(), reset_crop_region=True)
# Repeatedly using previous detection result to identify the region of
# interest and only croping that region to improve detection accuracy
for _ in range(inference_count - 1):
person = movenet.detect(input_tensor.numpy(),
reset_crop_region=False)
return person
Cloning into 'examples'... remote: Enumerating objects: 20141, done.[K remote: Counting objects: 100% (1961/1961), done.[K remote: Compressing objects: 100% (1055/1055), done.[K remote: Total 20141 (delta 909), reused 1584 (delta 595), pack-reused 18180[K Receiving objects: 100% (20141/20141), 33.15 MiB | 25.83 MiB/s, done. Resolving deltas: 100% (11003/11003), done.
Fonctions pour visualiser les résultats d'estimation de pose.
def draw_prediction_on_image(
image, person, crop_region=None, close_figure=True,
keep_input_size=False):
"""Draws the keypoint predictions on image.
Args:
image: An numpy array with shape [height, width, channel] representing the
pixel values of the input image.
person: A person entity returned from the MoveNet.SinglePose model.
close_figure: Whether to close the plt figure after the function returns.
keep_input_size: Whether to keep the size of the input image.
Returns:
An numpy array with shape [out_height, out_width, channel] representing the
image overlaid with keypoint predictions.
"""
# Draw the detection result on top of the image.
image_np = utils.visualize(image, [person])
# Plot the image with detection results.
height, width, channel = image.shape
aspect_ratio = float(width) / height
fig, ax = plt.subplots(figsize=(12 * aspect_ratio, 12))
im = ax.imshow(image_np)
if close_figure:
plt.close(fig)
if not keep_input_size:
image_np = utils.keep_aspect_ratio_resizer(image_np, (512, 512))
return image_np
Code pour charger les images, détecter les points de repère de pose et les enregistrer dans un fichier CSV
class MoveNetPreprocessor(object):
"""Helper class to preprocess pose sample images for classification."""
def __init__(self,
images_in_folder,
images_out_folder,
csvs_out_path):
"""Creates a preprocessor to detection pose from images and save as CSV.
Args:
images_in_folder: Path to the folder with the input images. It should
follow this structure:
yoga_poses
|__ downdog
|______ 00000128.jpg
|______ 00000181.bmp
|______ ...
|__ goddess
|______ 00000243.jpg
|______ 00000306.jpg
|______ ...
...
images_out_folder: Path to write the images overlay with detected
landmarks. These images are useful when you need to debug accuracy
issues.
csvs_out_path: Path to write the CSV containing the detected landmark
coordinates and label of each image that can be used to train a pose
classification model.
"""
self._images_in_folder = images_in_folder
self._images_out_folder = images_out_folder
self._csvs_out_path = csvs_out_path
self._messages = []
# Create a temp dir to store the pose CSVs per class
self._csvs_out_folder_per_class = tempfile.mkdtemp()
# Get list of pose classes and print image statistics
self._pose_class_names = sorted(
[n for n in os.listdir(self._images_in_folder) if not n.startswith('.')]
)
def process(self, per_pose_class_limit=None, detection_threshold=0.1):
"""Preprocesses images in the given folder.
Args:
per_pose_class_limit: Number of images to load. As preprocessing usually
takes time, this parameter can be specified to make the reduce of the
dataset for testing.
detection_threshold: Only keep images with all landmark confidence score
above this threshold.
"""
# Loop through the classes and preprocess its images
for pose_class_name in self._pose_class_names:
print('Preprocessing', pose_class_name, file=sys.stderr)
# Paths for the pose class.
images_in_folder = os.path.join(self._images_in_folder, pose_class_name)
images_out_folder = os.path.join(self._images_out_folder, pose_class_name)
csv_out_path = os.path.join(self._csvs_out_folder_per_class,
pose_class_name + '.csv')
if not os.path.exists(images_out_folder):
os.makedirs(images_out_folder)
# Detect landmarks in each image and write it to a CSV file
with open(csv_out_path, 'w') as csv_out_file:
csv_out_writer = csv.writer(csv_out_file,
delimiter=',',
quoting=csv.QUOTE_MINIMAL)
# Get list of images
image_names = sorted(
[n for n in os.listdir(images_in_folder) if not n.startswith('.')])
if per_pose_class_limit is not None:
image_names = image_names[:per_pose_class_limit]
valid_image_count = 0
# Detect pose landmarks from each image
for image_name in tqdm.tqdm(image_names):
image_path = os.path.join(images_in_folder, image_name)
try:
image = tf.io.read_file(image_path)
image = tf.io.decode_jpeg(image)
except:
self._messages.append('Skipped ' + image_path + '. Invalid image.')
continue
else:
image = tf.io.read_file(image_path)
image = tf.io.decode_jpeg(image)
image_height, image_width, channel = image.shape
# Skip images that isn't RGB because Movenet requires RGB images
if channel != 3:
self._messages.append('Skipped ' + image_path +
'. Image isn\'t in RGB format.')
continue
person = detect(image)
# Save landmarks if all landmarks were detected
min_landmark_score = min(
[keypoint.score for keypoint in person.keypoints])
should_keep_image = min_landmark_score >= detection_threshold
if not should_keep_image:
self._messages.append('Skipped ' + image_path +
'. No pose was confidentlly detected.')
continue
valid_image_count += 1
# Draw the prediction result on top of the image for debugging later
output_overlay = draw_prediction_on_image(
image.numpy().astype(np.uint8), person,
close_figure=True, keep_input_size=True)
# Write detection result into an image file
output_frame = cv2.cvtColor(output_overlay, cv2.COLOR_RGB2BGR)
cv2.imwrite(os.path.join(images_out_folder, image_name), output_frame)
# Get landmarks and scale it to the same size as the input image
pose_landmarks = np.array(
[[keypoint.coordinate.x, keypoint.coordinate.y, keypoint.score]
for keypoint in person.keypoints],
dtype=np.float32)
# Write the landmark coordinates to its per-class CSV file
coordinates = pose_landmarks.flatten().astype(np.str).tolist()
csv_out_writer.writerow([image_name] + coordinates)
if not valid_image_count:
raise RuntimeError(
'No valid images found for the "{}" class.'
.format(pose_class_name))
# Print the error message collected during preprocessing.
print('\n'.join(self._messages))
# Combine all per-class CSVs into a single output file
all_landmarks_df = self._all_landmarks_as_dataframe()
all_landmarks_df.to_csv(self._csvs_out_path, index=False)
def class_names(self):
"""List of classes found in the training dataset."""
return self._pose_class_names
def _all_landmarks_as_dataframe(self):
"""Merge all per-class CSVs into a single dataframe."""
total_df = None
for class_index, class_name in enumerate(self._pose_class_names):
csv_out_path = os.path.join(self._csvs_out_folder_per_class,
class_name + '.csv')
per_class_df = pd.read_csv(csv_out_path, header=None)
# Add the labels
per_class_df['class_no'] = [class_index]*len(per_class_df)
per_class_df['class_name'] = [class_name]*len(per_class_df)
# Append the folder name to the filename column (first column)
per_class_df[per_class_df.columns[0]] = (os.path.join(class_name, '')
+ per_class_df[per_class_df.columns[0]].astype(str))
if total_df is None:
# For the first class, assign its data to the total dataframe
total_df = per_class_df
else:
# Concatenate each class's data into the total dataframe
total_df = pd.concat([total_df, per_class_df], axis=0)
list_name = [[bodypart.name + '_x', bodypart.name + '_y',
bodypart.name + '_score'] for bodypart in BodyPart]
header_name = []
for columns_name in list_name:
header_name += columns_name
header_name = ['file_name'] + header_name
header_map = {total_df.columns[i]: header_name[i]
for i in range(len(header_name))}
total_df.rename(header_map, axis=1, inplace=True)
return total_df
(Facultatif) Extrait de code pour essayer la logique d'estimation de pose Movenet
test_image_url = "https://cdn.pixabay.com/photo/2017/03/03/17/30/yoga-2114512_960_720.jpg"
!wget -O /tmp/image.jpeg {test_image_url}
if len(test_image_url):
image = tf.io.read_file('/tmp/image.jpeg')
image = tf.io.decode_jpeg(image)
person = detect(image)
_ = draw_prediction_on_image(image.numpy(), person, crop_region=None,
close_figure=False, keep_input_size=True)
--2021-12-21 12:07:36-- https://cdn.pixabay.com/photo/2017/03/03/17/30/yoga-2114512_960_720.jpg Resolving cdn.pixabay.com (cdn.pixabay.com)... 104.18.20.183, 104.18.21.183, 2606:4700::6812:14b7, ... Connecting to cdn.pixabay.com (cdn.pixabay.com)|104.18.20.183|:443... connected. HTTP request sent, awaiting response... 200 OK Length: 28665 (28K) [image/jpeg] Saving to: ‘/tmp/image.jpeg’ /tmp/image.jpeg 100%[===================>] 27.99K --.-KB/s in 0s 2021-12-21 12:07:36 (111 MB/s) - ‘/tmp/image.jpeg’ saved [28665/28665]
Partie 1 : Prétraiter les images d'entrée
Parce que l'entrée pour notre classificateur pose est les points de repère de sortie du modèle MoveNet, nous devons générer nos données de formation en exécutant les images marquées par MoveNet puis capturer toutes les données historiques et les étiquettes de vérité au sol dans un fichier CSV.
L'ensemble de données que nous avons fourni pour ce didacticiel est un ensemble de données de pose de yoga généré par CG. Il contient des images de plusieurs modèles générés par CG faisant 5 poses de yoga différentes. Le répertoire est déjà divisé en un train de jeu de données et un test de jeu de données.
Donc , dans cette section, nous téléchargeons le jeu de données de yoga et courir par MoveNet afin que nous puissions capturer tous les points de repère dans un fichier CSV ... Cependant, il faut environ 15 minutes pour alimenter notre ensemble de données de yoga à MoveNet et générons ce fichier CSV . Donc , comme alternative, vous pouvez télécharger un fichier CSV préexistant pour l'ensemble de données de yoga en définissant is_skip_step_1 paramètre ci - dessous pour True. De cette façon, vous ignorerez cette étape et téléchargerez à la place le même fichier CSV qui sera créé lors de cette étape de prétraitement.
D'autre part, si vous voulez former le classificateur de pose avec votre propre jeu de données d'image, vous devez télécharger vos images et exécuter cette étape de pré - traitement (congé is_skip_step_1 de False) -Suivre les instructions ci - dessous pour télécharger votre propre jeu de données pose.
is_skip_step_1 = False
(Facultatif) Téléchargez votre propre jeu de données de pose
use_custom_dataset = False
dataset_is_split = False
Si vous souhaitez entraîner le classificateur de poses avec vos propres poses étiquetées (il peut s'agir de n'importe quelle pose, pas seulement de poses de yoga), suivez ces étapes :
Réglez ci - dessus
use_custom_datasetoption True.Préparez un fichier d'archive (ZIP, TAR ou autre) qui inclut un dossier avec votre jeu de données d'images. Le dossier doit inclure des images triées de vos poses comme suit.
Si vous avez déjà diviser votre ensemble de données en ensembles de train et de test, puis définissez
dataset_is_splitsur True. C'est-à-dire que votre dossier d'images doit inclure des répertoires « train » et « test » comme celui-ci :yoga_poses/ |__ train/ |__ downdog/ |______ 00000128.jpg |______ ... |__ test/ |__ downdog/ |______ 00000181.jpg |______ ...Ou, si votre ensemble de données est divisé PAS encore, puis réglez
dataset_is_splitsur False et nous allons le diviser la base d'une fraction de fraction spécifiée. C'est-à-dire que votre dossier d'images téléchargées devrait ressembler à ceci :yoga_poses/ |__ downdog/ |______ 00000128.jpg |______ 00000181.jpg |______ ... |__ goddess/ |______ 00000243.jpg |______ 00000306.jpg |______ ...Cliquez sur l'onglet Fichiers sur le (icône de dossier) à gauche, puis cliquez sur Télécharger pour le stockage de session (icône du fichier).
Sélectionnez votre fichier d'archive et attendez la fin du téléchargement avant de continuer.
Modifiez le bloc de code suivant pour spécifier le nom de votre fichier d'archive et du répertoire d'images. (Par défaut, nous attendons un fichier ZIP, vous devrez donc également modifier cette partie si votre archive est dans un autre format.)
Exécutez maintenant le reste du bloc-notes.
import os
import random
import shutil
def split_into_train_test(images_origin, images_dest, test_split):
"""Splits a directory of sorted images into training and test sets.
Args:
images_origin: Path to the directory with your images. This directory
must include subdirectories for each of your labeled classes. For example:
yoga_poses/
|__ downdog/
|______ 00000128.jpg
|______ 00000181.jpg
|______ ...
|__ goddess/
|______ 00000243.jpg
|______ 00000306.jpg
|______ ...
...
images_dest: Path to a directory where you want the split dataset to be
saved. The results looks like this:
split_yoga_poses/
|__ train/
|__ downdog/
|______ 00000128.jpg
|______ ...
|__ test/
|__ downdog/
|______ 00000181.jpg
|______ ...
test_split: Fraction of data to reserve for test (float between 0 and 1).
"""
_, dirs, _ = next(os.walk(images_origin))
TRAIN_DIR = os.path.join(images_dest, 'train')
TEST_DIR = os.path.join(images_dest, 'test')
os.makedirs(TRAIN_DIR, exist_ok=True)
os.makedirs(TEST_DIR, exist_ok=True)
for dir in dirs:
# Get all filenames for this dir, filtered by filetype
filenames = os.listdir(os.path.join(images_origin, dir))
filenames = [os.path.join(images_origin, dir, f) for f in filenames if (
f.endswith('.png') or f.endswith('.jpg') or f.endswith('.jpeg') or f.endswith('.bmp'))]
# Shuffle the files, deterministically
filenames.sort()
random.seed(42)
random.shuffle(filenames)
# Divide them into train/test dirs
os.makedirs(os.path.join(TEST_DIR, dir), exist_ok=True)
os.makedirs(os.path.join(TRAIN_DIR, dir), exist_ok=True)
test_count = int(len(filenames) * test_split)
for i, file in enumerate(filenames):
if i < test_count:
destination = os.path.join(TEST_DIR, dir, os.path.split(file)[1])
else:
destination = os.path.join(TRAIN_DIR, dir, os.path.split(file)[1])
shutil.copyfile(file, destination)
print(f'Moved {test_count} of {len(filenames)} from class "{dir}" into test.')
print(f'Your split dataset is in "{images_dest}"')
if use_custom_dataset:
# ATTENTION:
# You must edit these two lines to match your archive and images folder name:
# !tar -xf YOUR_DATASET_ARCHIVE_NAME.tar
!unzip -q YOUR_DATASET_ARCHIVE_NAME.zip
dataset_in = 'YOUR_DATASET_DIR_NAME'
# You can leave the rest alone:
if not os.path.isdir(dataset_in):
raise Exception("dataset_in is not a valid directory")
if dataset_is_split:
IMAGES_ROOT = dataset_in
else:
dataset_out = 'split_' + dataset_in
split_into_train_test(dataset_in, dataset_out, test_split=0.2)
IMAGES_ROOT = dataset_out
Télécharger le jeu de données de yoga
if not is_skip_step_1 and not use_custom_dataset:
!wget -O yoga_poses.zip http://download.tensorflow.org/data/pose_classification/yoga_poses.zip
!unzip -q yoga_poses.zip -d yoga_cg
IMAGES_ROOT = "yoga_cg"
--2021-12-21 12:07:46-- http://download.tensorflow.org/data/pose_classification/yoga_poses.zip Resolving download.tensorflow.org (download.tensorflow.org)... 172.217.218.128, 2a00:1450:4013:c08::80 Connecting to download.tensorflow.org (download.tensorflow.org)|172.217.218.128|:80... connected. HTTP request sent, awaiting response... 200 OK Length: 102517581 (98M) [application/zip] Saving to: ‘yoga_poses.zip’ yoga_poses.zip 100%[===================>] 97.77M 76.7MB/s in 1.3s 2021-12-21 12:07:48 (76.7 MB/s) - ‘yoga_poses.zip’ saved [102517581/102517581]
Préprocesser TRAIN ensemble de données
if not is_skip_step_1:
images_in_train_folder = os.path.join(IMAGES_ROOT, 'train')
images_out_train_folder = 'poses_images_out_train'
csvs_out_train_path = 'train_data.csv'
preprocessor = MoveNetPreprocessor(
images_in_folder=images_in_train_folder,
images_out_folder=images_out_train_folder,
csvs_out_path=csvs_out_train_path,
)
preprocessor.process(per_pose_class_limit=None)
Preprocessing chair 0%| | 0/200 [00:00<?, ?it/s]/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/ipykernel_launcher.py:128: DeprecationWarning: `np.str` is a deprecated alias for the builtin `str`. To silence this warning, use `str` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.str_` here. Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations 100%|██████████| 200/200 [00:32<00:00, 6.10it/s] Preprocessing cobra 100%|██████████| 200/200 [00:31<00:00, 6.27it/s] Preprocessing dog 100%|██████████| 200/200 [00:32<00:00, 6.15it/s] Preprocessing tree 100%|██████████| 200/200 [00:33<00:00, 6.00it/s] Preprocessing warrior 100%|██████████| 200/200 [00:30<00:00, 6.54it/s] Skipped yoga_cg/train/chair/girl3_chair091.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair092.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair093.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair094.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair096.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair097.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair099.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair100.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair104.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair106.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair110.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair114.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair115.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair118.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair122.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair123.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair124.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair125.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair131.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair132.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair133.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair134.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair136.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair138.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair139.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/girl3_chair142.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/guy2_chair089.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/guy2_chair136.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/guy2_chair140.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/guy2_chair143.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/guy2_chair144.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/guy2_chair145.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/chair/guy2_chair146.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra026.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra029.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra030.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra038.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra040.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra041.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra048.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra050.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra051.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra055.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra059.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra061.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra068.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra070.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra081.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra087.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra088.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra089.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra090.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra091.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra092.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra093.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra094.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra096.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra099.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra102.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra110.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra112.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra115.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra119.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra122.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra128.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra129.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra136.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl1_cobra140.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl2_cobra029.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl2_cobra046.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl2_cobra050.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl2_cobra053.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl2_cobra108.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl2_cobra117.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl2_cobra129.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl2_cobra133.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl2_cobra136.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl2_cobra140.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra028.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra030.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra032.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra039.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra040.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra051.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra052.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra058.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra062.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra068.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra072.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra076.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra078.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra079.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra082.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra088.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra092.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra097.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra099.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra107.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra129.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra130.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra132.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra134.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/girl3_cobra138.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra034.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra042.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra043.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra047.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra053.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra065.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra077.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra078.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra080.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra081.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra084.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra088.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra089.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra102.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra105.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra108.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/cobra/guy2_cobra139.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl1_dog027.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl1_dog028.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl1_dog030.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl1_dog032.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog075.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog080.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog083.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog085.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog087.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog088.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog090.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog091.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog093.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog094.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog095.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog099.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog100.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog101.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog103.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog104.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog105.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog107.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl2_dog111.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog025.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog026.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog027.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog028.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog031.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog033.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog035.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog037.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog040.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog041.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog047.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog052.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog062.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog072.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog074.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog075.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog077.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog081.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog082.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog086.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog088.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog090.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog092.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog094.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog095.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog096.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog100.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog102.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog103.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog104.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog106.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog107.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/girl3_dog111.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/guy1_dog070.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/guy1_dog076.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/guy2_dog070.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/guy2_dog071.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/dog/guy2_dog082.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/girl2_tree119.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/girl2_tree122.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/girl2_tree161.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/girl2_tree163.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy1_tree139.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy1_tree140.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy1_tree141.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy1_tree143.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy2_tree085.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy2_tree086.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy2_tree087.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy2_tree088.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy2_tree090.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy2_tree092.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy2_tree145.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/tree/guy2_tree147.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior049.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior053.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior064.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior066.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior067.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior072.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior075.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior077.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior080.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior083.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior084.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior087.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior089.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior093.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior094.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior095.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior098.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior099.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior100.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior103.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior108.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior109.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior111.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior112.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior113.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior114.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior116.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl1_warrior117.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior047.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior049.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior050.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior052.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior057.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior058.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior063.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior068.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior079.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior083.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior085.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior088.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior092.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior096.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior097.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior102.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior106.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl2_warrior108.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior042.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior043.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior047.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior049.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior051.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior054.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior056.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior057.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior061.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior066.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior067.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior073.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior074.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior075.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior079.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior087.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior089.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior090.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior091.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior092.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior094.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior095.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior096.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior100.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior103.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior107.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior115.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior117.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior134.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior140.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/girl3_warrior143.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior043.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior048.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior051.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior052.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior055.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior057.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior062.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior068.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior069.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior073.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior076.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior077.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior080.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior081.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior082.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior088.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior091.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior092.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior093.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior094.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior097.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior118.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior120.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior121.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior124.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior125.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior126.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior131.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior134.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior135.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior138.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior143.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior145.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy1_warrior148.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior051.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior086.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior111.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior118.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior122.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior129.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior131.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior135.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior137.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior139.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior145.jpg. No pose was confidentlly detected. Skipped yoga_cg/train/warrior/guy2_warrior148.jpg. No pose was confidentlly detected.
Préprocesser TEST ensemble de données
if not is_skip_step_1:
images_in_test_folder = os.path.join(IMAGES_ROOT, 'test')
images_out_test_folder = 'poses_images_out_test'
csvs_out_test_path = 'test_data.csv'
preprocessor = MoveNetPreprocessor(
images_in_folder=images_in_test_folder,
images_out_folder=images_out_test_folder,
csvs_out_path=csvs_out_test_path,
)
preprocessor.process(per_pose_class_limit=None)
Preprocessing chair 0%| | 0/84 [00:00<?, ?it/s]/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/ipykernel_launcher.py:128: DeprecationWarning: `np.str` is a deprecated alias for the builtin `str`. To silence this warning, use `str` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.str_` here. Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations 100%|██████████| 84/84 [00:15<00:00, 5.51it/s] Preprocessing cobra 100%|██████████| 116/116 [00:19<00:00, 6.10it/s] Preprocessing dog 100%|██████████| 90/90 [00:14<00:00, 6.03it/s] Preprocessing tree 100%|██████████| 96/96 [00:16<00:00, 5.98it/s] Preprocessing warrior 100%|██████████| 109/109 [00:17<00:00, 6.38it/s] Skipped yoga_cg/test/cobra/guy3_cobra048.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra050.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra051.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra052.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra053.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra054.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra055.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra056.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra057.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra058.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra059.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra060.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra062.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra069.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra075.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra077.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra081.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra124.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra131.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra132.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra134.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra135.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/cobra/guy3_cobra136.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/dog/guy3_dog025.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/dog/guy3_dog026.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/dog/guy3_dog036.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/dog/guy3_dog042.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/dog/guy3_dog106.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/dog/guy3_dog108.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior042.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior043.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior044.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior045.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior046.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior047.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior048.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior050.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior051.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior052.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior053.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior054.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior055.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior056.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior059.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior060.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior062.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior063.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior065.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior066.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior068.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior070.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior071.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior072.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior073.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior074.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior075.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior076.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior077.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior079.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior080.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior081.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior082.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior083.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior084.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior085.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior086.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior087.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior088.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior089.jpg. No pose was confidentlly detected. Skipped yoga_cg/test/warrior/guy3_warrior137.jpg. No pose was confidentlly detected.
Partie 2 : Entraînez un modèle de classification de pose qui prend les coordonnées du point de repère en entrée et génère les étiquettes prédites.
Vous allez créer un modèle TensorFlow qui prend les coordonnées du point de repère et prédit la classe de pose que la personne dans l'image d'entrée effectue. Le modèle se compose de deux sous-modèles :
- Le sous-modèle 1 calcule un plongement de pose (alias vecteur de caractéristiques) à partir des coordonnées de point de repère détectées.
- 2 flux sous - modèle posent l' intégration par plusieurs
Densecouche pour prédire la classe de pose.
Vous entraînerez ensuite le modèle en fonction de l'ensemble de données prétraité dans la partie 1.
(Facultatif) Téléchargez l'ensemble de données prétraité si vous n'avez pas exécuté la partie 1
# Download the preprocessed CSV files which are the same as the output of step 1
if is_skip_step_1:
!wget -O train_data.csv http://download.tensorflow.org/data/pose_classification/yoga_train_data.csv
!wget -O test_data.csv http://download.tensorflow.org/data/pose_classification/yoga_test_data.csv
csvs_out_train_path = 'train_data.csv'
csvs_out_test_path = 'test_data.csv'
is_skipped_step_1 = True
Charger les CSVs en prétraitées TRAIN et TEST jeux de données.
def load_pose_landmarks(csv_path):
"""Loads a CSV created by MoveNetPreprocessor.
Returns:
X: Detected landmark coordinates and scores of shape (N, 17 * 3)
y: Ground truth labels of shape (N, label_count)
classes: The list of all class names found in the dataset
dataframe: The CSV loaded as a Pandas dataframe features (X) and ground
truth labels (y) to use later to train a pose classification model.
"""
# Load the CSV file
dataframe = pd.read_csv(csv_path)
df_to_process = dataframe.copy()
# Drop the file_name columns as you don't need it during training.
df_to_process.drop(columns=['file_name'], inplace=True)
# Extract the list of class names
classes = df_to_process.pop('class_name').unique()
# Extract the labels
y = df_to_process.pop('class_no')
# Convert the input features and labels into the correct format for training.
X = df_to_process.astype('float64')
y = keras.utils.to_categorical(y)
return X, y, classes, dataframe
Charger et diviser l'original TRAIN ensemble de données en TRAIN (85% des données) et VALIDATE (les 15% restants).
# Load the train data
X, y, class_names, _ = load_pose_landmarks(csvs_out_train_path)
# Split training data (X, y) into (X_train, y_train) and (X_val, y_val)
X_train, X_val, y_train, y_val = train_test_split(X, y,
test_size=0.15)
# Load the test data
X_test, y_test, _, df_test = load_pose_landmarks(csvs_out_test_path)
Définir des fonctions pour convertir les points de repère de la pose en une intégration de pose (alias vecteur de caractéristiques) pour la classification des poses
Ensuite, convertissez les coordonnées du point de repère en un vecteur d'entité en :
- Déplacer le centre de la pose vers l'origine.
- Redimensionner la pose pour que la taille de la pose devienne 1
- Aplatir ces coordonnées dans un vecteur de caractéristiques
Utilisez ensuite ce vecteur de fonctionnalités pour entraîner un classificateur de pose basé sur un réseau de neurones.
def get_center_point(landmarks, left_bodypart, right_bodypart):
"""Calculates the center point of the two given landmarks."""
left = tf.gather(landmarks, left_bodypart.value, axis=1)
right = tf.gather(landmarks, right_bodypart.value, axis=1)
center = left * 0.5 + right * 0.5
return center
def get_pose_size(landmarks, torso_size_multiplier=2.5):
"""Calculates pose size.
It is the maximum of two values:
* Torso size multiplied by `torso_size_multiplier`
* Maximum distance from pose center to any pose landmark
"""
# Hips center
hips_center = get_center_point(landmarks, BodyPart.LEFT_HIP,
BodyPart.RIGHT_HIP)
# Shoulders center
shoulders_center = get_center_point(landmarks, BodyPart.LEFT_SHOULDER,
BodyPart.RIGHT_SHOULDER)
# Torso size as the minimum body size
torso_size = tf.linalg.norm(shoulders_center - hips_center)
# Pose center
pose_center_new = get_center_point(landmarks, BodyPart.LEFT_HIP,
BodyPart.RIGHT_HIP)
pose_center_new = tf.expand_dims(pose_center_new, axis=1)
# Broadcast the pose center to the same size as the landmark vector to
# perform substraction
pose_center_new = tf.broadcast_to(pose_center_new,
[tf.size(landmarks) // (17*2), 17, 2])
# Dist to pose center
d = tf.gather(landmarks - pose_center_new, 0, axis=0,
name="dist_to_pose_center")
# Max dist to pose center
max_dist = tf.reduce_max(tf.linalg.norm(d, axis=0))
# Normalize scale
pose_size = tf.maximum(torso_size * torso_size_multiplier, max_dist)
return pose_size
def normalize_pose_landmarks(landmarks):
"""Normalizes the landmarks translation by moving the pose center to (0,0) and
scaling it to a constant pose size.
"""
# Move landmarks so that the pose center becomes (0,0)
pose_center = get_center_point(landmarks, BodyPart.LEFT_HIP,
BodyPart.RIGHT_HIP)
pose_center = tf.expand_dims(pose_center, axis=1)
# Broadcast the pose center to the same size as the landmark vector to perform
# substraction
pose_center = tf.broadcast_to(pose_center,
[tf.size(landmarks) // (17*2), 17, 2])
landmarks = landmarks - pose_center
# Scale the landmarks to a constant pose size
pose_size = get_pose_size(landmarks)
landmarks /= pose_size
return landmarks
def landmarks_to_embedding(landmarks_and_scores):
"""Converts the input landmarks into a pose embedding."""
# Reshape the flat input into a matrix with shape=(17, 3)
reshaped_inputs = keras.layers.Reshape((17, 3))(landmarks_and_scores)
# Normalize landmarks 2D
landmarks = normalize_pose_landmarks(reshaped_inputs[:, :, :2])
# Flatten the normalized landmark coordinates into a vector
embedding = keras.layers.Flatten()(landmarks)
return embedding
Définir un modèle Keras pour la classification des poses
Notre modèle Keras prend les repères de pose détectés, puis calcule l'intégration de la pose et prédit la classe de pose.
# Define the model
inputs = tf.keras.Input(shape=(51))
embedding = landmarks_to_embedding(inputs)
layer = keras.layers.Dense(128, activation=tf.nn.relu6)(embedding)
layer = keras.layers.Dropout(0.5)(layer)
layer = keras.layers.Dense(64, activation=tf.nn.relu6)(layer)
layer = keras.layers.Dropout(0.5)(layer)
outputs = keras.layers.Dense(len(class_names), activation="softmax")(layer)
model = keras.Model(inputs, outputs)
model.summary()
Model: "model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 51)] 0 []
reshape (Reshape) (None, 17, 3) 0 ['input_1[0][0]']
tf.__operators__.getitem (Slic (None, 17, 2) 0 ['reshape[0][0]']
ingOpLambda)
tf.compat.v1.gather (TFOpLambd (None, 2) 0 ['tf.__operators__.getitem[0][0]'
a) ]
tf.compat.v1.gather_1 (TFOpLam (None, 2) 0 ['tf.__operators__.getitem[0][0]'
bda) ]
tf.math.multiply (TFOpLambda) (None, 2) 0 ['tf.compat.v1.gather[0][0]']
tf.math.multiply_1 (TFOpLambda (None, 2) 0 ['tf.compat.v1.gather_1[0][0]']
)
tf.__operators__.add (TFOpLamb (None, 2) 0 ['tf.math.multiply[0][0]',
da) 'tf.math.multiply_1[0][0]']
tf.compat.v1.size (TFOpLambda) () 0 ['tf.__operators__.getitem[0][0]'
]
tf.expand_dims (TFOpLambda) (None, 1, 2) 0 ['tf.__operators__.add[0][0]']
tf.compat.v1.floor_div (TFOpLa () 0 ['tf.compat.v1.size[0][0]']
mbda)
tf.broadcast_to (TFOpLambda) (None, 17, 2) 0 ['tf.expand_dims[0][0]',
'tf.compat.v1.floor_div[0][0]']
tf.math.subtract (TFOpLambda) (None, 17, 2) 0 ['tf.__operators__.getitem[0][0]'
, 'tf.broadcast_to[0][0]']
tf.compat.v1.gather_6 (TFOpLam (None, 2) 0 ['tf.math.subtract[0][0]']
bda)
tf.compat.v1.gather_7 (TFOpLam (None, 2) 0 ['tf.math.subtract[0][0]']
bda)
tf.math.multiply_6 (TFOpLambda (None, 2) 0 ['tf.compat.v1.gather_6[0][0]']
)
tf.math.multiply_7 (TFOpLambda (None, 2) 0 ['tf.compat.v1.gather_7[0][0]']
)
tf.__operators__.add_3 (TFOpLa (None, 2) 0 ['tf.math.multiply_6[0][0]',
mbda) 'tf.math.multiply_7[0][0]']
tf.compat.v1.size_1 (TFOpLambd () 0 ['tf.math.subtract[0][0]']
a)
tf.compat.v1.gather_4 (TFOpLam (None, 2) 0 ['tf.math.subtract[0][0]']
bda)
tf.compat.v1.gather_5 (TFOpLam (None, 2) 0 ['tf.math.subtract[0][0]']
bda)
tf.compat.v1.gather_2 (TFOpLam (None, 2) 0 ['tf.math.subtract[0][0]']
bda)
tf.compat.v1.gather_3 (TFOpLam (None, 2) 0 ['tf.math.subtract[0][0]']
bda)
tf.expand_dims_1 (TFOpLambda) (None, 1, 2) 0 ['tf.__operators__.add_3[0][0]']
tf.compat.v1.floor_div_1 (TFOp () 0 ['tf.compat.v1.size_1[0][0]']
Lambda)
tf.math.multiply_4 (TFOpLambda (None, 2) 0 ['tf.compat.v1.gather_4[0][0]']
)
tf.math.multiply_5 (TFOpLambda (None, 2) 0 ['tf.compat.v1.gather_5[0][0]']
)
tf.math.multiply_2 (TFOpLambda (None, 2) 0 ['tf.compat.v1.gather_2[0][0]']
)
tf.math.multiply_3 (TFOpLambda (None, 2) 0 ['tf.compat.v1.gather_3[0][0]']
)
tf.broadcast_to_1 (TFOpLambda) (None, 17, 2) 0 ['tf.expand_dims_1[0][0]',
'tf.compat.v1.floor_div_1[0][0]'
]
tf.__operators__.add_2 (TFOpLa (None, 2) 0 ['tf.math.multiply_4[0][0]',
mbda) 'tf.math.multiply_5[0][0]']
tf.__operators__.add_1 (TFOpLa (None, 2) 0 ['tf.math.multiply_2[0][0]',
mbda) 'tf.math.multiply_3[0][0]']
tf.math.subtract_2 (TFOpLambda (None, 17, 2) 0 ['tf.math.subtract[0][0]',
) 'tf.broadcast_to_1[0][0]']
tf.math.subtract_1 (TFOpLambda (None, 2) 0 ['tf.__operators__.add_2[0][0]',
) 'tf.__operators__.add_1[0][0]']
tf.compat.v1.gather_8 (TFOpLam (17, 2) 0 ['tf.math.subtract_2[0][0]']
bda)
tf.compat.v1.norm (TFOpLambda) () 0 ['tf.math.subtract_1[0][0]']
tf.compat.v1.norm_1 (TFOpLambd (2,) 0 ['tf.compat.v1.gather_8[0][0]']
a)
tf.math.multiply_8 (TFOpLambda () 0 ['tf.compat.v1.norm[0][0]']
)
tf.math.reduce_max (TFOpLambda () 0 ['tf.compat.v1.norm_1[0][0]']
)
tf.math.maximum (TFOpLambda) () 0 ['tf.math.multiply_8[0][0]',
'tf.math.reduce_max[0][0]']
tf.math.truediv (TFOpLambda) (None, 17, 2) 0 ['tf.math.subtract[0][0]',
'tf.math.maximum[0][0]']
flatten (Flatten) (None, 34) 0 ['tf.math.truediv[0][0]']
dense (Dense) (None, 128) 4480 ['flatten[0][0]']
dropout (Dropout) (None, 128) 0 ['dense[0][0]']
dense_1 (Dense) (None, 64) 8256 ['dropout[0][0]']
dropout_1 (Dropout) (None, 64) 0 ['dense_1[0][0]']
dense_2 (Dense) (None, 5) 325 ['dropout_1[0][0]']
==================================================================================================
Total params: 13,061
Trainable params: 13,061
Non-trainable params: 0
__________________________________________________________________________________________________
model.compile(
optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy']
)
# Add a checkpoint callback to store the checkpoint that has the highest
# validation accuracy.
checkpoint_path = "weights.best.hdf5"
checkpoint = keras.callbacks.ModelCheckpoint(checkpoint_path,
monitor='val_accuracy',
verbose=1,
save_best_only=True,
mode='max')
earlystopping = keras.callbacks.EarlyStopping(monitor='val_accuracy',
patience=20)
# Start training
history = model.fit(X_train, y_train,
epochs=200,
batch_size=16,
validation_data=(X_val, y_val),
callbacks=[checkpoint, earlystopping])
Epoch 1/200 19/37 [==============>...............] - ETA: 0s - loss: 1.5703 - accuracy: 0.3684 Epoch 00001: val_accuracy improved from -inf to 0.64706, saving model to weights.best.hdf5 37/37 [==============================] - 1s 11ms/step - loss: 1.5090 - accuracy: 0.4602 - val_loss: 1.3352 - val_accuracy: 0.6471 Epoch 2/200 20/37 [===============>..............] - ETA: 0s - loss: 1.3372 - accuracy: 0.4844 Epoch 00002: val_accuracy improved from 0.64706 to 0.67647, saving model to weights.best.hdf5 37/37 [==============================] - 0s 4ms/step - loss: 1.2375 - accuracy: 0.5190 - val_loss: 1.0193 - val_accuracy: 0.6765 Epoch 3/200 20/37 [===============>..............] - ETA: 0s - loss: 1.0596 - accuracy: 0.5469 Epoch 00003: val_accuracy improved from 0.67647 to 0.75490, saving model to weights.best.hdf5 37/37 [==============================] - 0s 4ms/step - loss: 1.0096 - accuracy: 0.5796 - val_loss: 0.8397 - val_accuracy: 0.7549 Epoch 4/200 21/37 [================>.............] - ETA: 0s - loss: 0.8922 - accuracy: 0.6220 Epoch 00004: val_accuracy improved from 0.75490 to 0.81373, saving model to weights.best.hdf5 37/37 [==============================] - 0s 4ms/step - loss: 0.8798 - accuracy: 0.6349 - val_loss: 0.7103 - val_accuracy: 0.8137 Epoch 5/200 20/37 [===============>..............] - ETA: 0s - loss: 0.7895 - accuracy: 0.6875 Epoch 00005: val_accuracy improved from 0.81373 to 0.82353, saving model to weights.best.hdf5 37/37 [==============================] - 0s 4ms/step - loss: 0.7810 - accuracy: 0.6903 - val_loss: 0.6120 - val_accuracy: 0.8235 Epoch 6/200 20/37 [===============>..............] - ETA: 0s - loss: 0.7324 - accuracy: 0.7250 Epoch 00006: val_accuracy improved from 0.82353 to 0.92157, saving model to weights.best.hdf5 37/37 [==============================] - 0s 4ms/step - loss: 0.7263 - accuracy: 0.7093 - val_loss: 0.5297 - val_accuracy: 0.9216 Epoch 7/200 19/37 [==============>...............] - ETA: 0s - loss: 0.6852 - accuracy: 0.7467 Epoch 00007: val_accuracy did not improve from 0.92157 37/37 [==============================] - 0s 4ms/step - loss: 0.6450 - accuracy: 0.7595 - val_loss: 0.4635 - val_accuracy: 0.8922 Epoch 8/200 20/37 [===============>..............] - ETA: 0s - loss: 0.6007 - accuracy: 0.7719 Epoch 00008: val_accuracy did not improve from 0.92157 37/37 [==============================] - 0s 4ms/step - loss: 0.5751 - accuracy: 0.7837 - val_loss: 0.4012 - val_accuracy: 0.9216 Epoch 9/200 20/37 [===============>..............] - ETA: 0s - loss: 0.5358 - accuracy: 0.8125 Epoch 00009: val_accuracy improved from 0.92157 to 0.93137, saving model to weights.best.hdf5 37/37 [==============================] - 0s 4ms/step - loss: 0.5272 - accuracy: 0.8097 - val_loss: 0.3547 - val_accuracy: 0.9314 Epoch 10/200 20/37 [===============>..............] - ETA: 0s - loss: 0.5200 - accuracy: 0.8094 Epoch 00010: val_accuracy improved from 0.93137 to 0.98039, saving model to weights.best.hdf5 37/37 [==============================] - 0s 5ms/step - loss: 0.5051 - accuracy: 0.8218 - val_loss: 0.3014 - val_accuracy: 0.9804 Epoch 11/200 19/37 [==============>...............] - ETA: 0s - loss: 0.4413 - accuracy: 0.8322 Epoch 00011: val_accuracy did not improve from 0.98039 37/37 [==============================] - 0s 4ms/step - loss: 0.4509 - accuracy: 0.8374 - val_loss: 0.2786 - val_accuracy: 0.9706 Epoch 12/200 20/37 [===============>..............] - ETA: 0s - loss: 0.4323 - accuracy: 0.8156 Epoch 00012: val_accuracy improved from 0.98039 to 0.99020, saving model to weights.best.hdf5 37/37 [==============================] - 0s 5ms/step - loss: 0.4377 - accuracy: 0.8253 - val_loss: 0.2440 - val_accuracy: 0.9902 Epoch 13/200 20/37 [===============>..............] - ETA: 0s - loss: 0.4037 - accuracy: 0.8719 Epoch 00013: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.4187 - accuracy: 0.8668 - val_loss: 0.2109 - val_accuracy: 0.9804 Epoch 14/200 20/37 [===============>..............] - ETA: 0s - loss: 0.3664 - accuracy: 0.8813 Epoch 00014: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.3733 - accuracy: 0.8772 - val_loss: 0.2030 - val_accuracy: 0.9804 Epoch 15/200 20/37 [===============>..............] - ETA: 0s - loss: 0.3708 - accuracy: 0.8781 Epoch 00015: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.3684 - accuracy: 0.8754 - val_loss: 0.1765 - val_accuracy: 0.9902 Epoch 16/200 21/37 [================>.............] - ETA: 0s - loss: 0.3238 - accuracy: 0.9137 Epoch 00016: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.3213 - accuracy: 0.9100 - val_loss: 0.1662 - val_accuracy: 0.9804 Epoch 17/200 20/37 [===============>..............] - ETA: 0s - loss: 0.2739 - accuracy: 0.9281 Epoch 00017: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.3015 - accuracy: 0.9100 - val_loss: 0.1423 - val_accuracy: 0.9804 Epoch 18/200 20/37 [===============>..............] - ETA: 0s - loss: 0.3076 - accuracy: 0.9062 Epoch 00018: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.3022 - accuracy: 0.9048 - val_loss: 0.1407 - val_accuracy: 0.9804 Epoch 19/200 20/37 [===============>..............] - ETA: 0s - loss: 0.2719 - accuracy: 0.9250 Epoch 00019: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.2697 - accuracy: 0.9291 - val_loss: 0.1191 - val_accuracy: 0.9902 Epoch 20/200 20/37 [===============>..............] - ETA: 0s - loss: 0.2960 - accuracy: 0.9031 Epoch 00020: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.2775 - accuracy: 0.9100 - val_loss: 0.1120 - val_accuracy: 0.9902 Epoch 21/200 20/37 [===============>..............] - ETA: 0s - loss: 0.2590 - accuracy: 0.9250 Epoch 00021: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.2537 - accuracy: 0.9273 - val_loss: 0.1022 - val_accuracy: 0.9902 Epoch 22/200 20/37 [===============>..............] - ETA: 0s - loss: 0.2504 - accuracy: 0.9344 Epoch 00022: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.2661 - accuracy: 0.9204 - val_loss: 0.0976 - val_accuracy: 0.9902 Epoch 23/200 20/37 [===============>..............] - ETA: 0s - loss: 0.2384 - accuracy: 0.9156 Epoch 00023: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.2182 - accuracy: 0.9308 - val_loss: 0.0944 - val_accuracy: 0.9902 Epoch 24/200 20/37 [===============>..............] - ETA: 0s - loss: 0.2157 - accuracy: 0.9375 Epoch 00024: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.2031 - accuracy: 0.9412 - val_loss: 0.0844 - val_accuracy: 0.9902 Epoch 25/200 20/37 [===============>..............] - ETA: 0s - loss: 0.1944 - accuracy: 0.9469 Epoch 00025: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.2080 - accuracy: 0.9343 - val_loss: 0.0811 - val_accuracy: 0.9902 Epoch 26/200 20/37 [===============>..............] - ETA: 0s - loss: 0.2232 - accuracy: 0.9312 Epoch 00026: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.2033 - accuracy: 0.9394 - val_loss: 0.0703 - val_accuracy: 0.9902 Epoch 27/200 20/37 [===============>..............] - ETA: 0s - loss: 0.2120 - accuracy: 0.9281 Epoch 00027: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.1845 - accuracy: 0.9481 - val_loss: 0.0708 - val_accuracy: 0.9902 Epoch 28/200 20/37 [===============>..............] - ETA: 0s - loss: 0.2696 - accuracy: 0.9156 Epoch 00028: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.2355 - accuracy: 0.9273 - val_loss: 0.0679 - val_accuracy: 0.9902 Epoch 29/200 20/37 [===============>..............] - ETA: 0s - loss: 0.1794 - accuracy: 0.9531 Epoch 00029: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.1938 - accuracy: 0.9498 - val_loss: 0.0623 - val_accuracy: 0.9902 Epoch 30/200 20/37 [===============>..............] - ETA: 0s - loss: 0.1831 - accuracy: 0.9406 Epoch 00030: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.1758 - accuracy: 0.9498 - val_loss: 0.0599 - val_accuracy: 0.9902 Epoch 31/200 20/37 [===============>..............] - ETA: 0s - loss: 0.1967 - accuracy: 0.9375 Epoch 00031: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.1724 - accuracy: 0.9516 - val_loss: 0.0565 - val_accuracy: 0.9902 Epoch 32/200 20/37 [===============>..............] - ETA: 0s - loss: 0.1868 - accuracy: 0.9219 Epoch 00032: val_accuracy did not improve from 0.99020 37/37 [==============================] - 0s 4ms/step - loss: 0.1676 - accuracy: 0.9360 - val_loss: 0.0503 - val_accuracy: 0.9902
# Visualize the training history to see whether you're overfitting.
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('Model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['TRAIN', 'VAL'], loc='lower right')
plt.show()
# Evaluate the model using the TEST dataset
loss, accuracy = model.evaluate(X_test, y_test)
14/14 [==============================] - 0s 2ms/step - loss: 0.0612 - accuracy: 0.9976
Dessinez la matrice de confusion pour mieux comprendre les performances du modèle
def plot_confusion_matrix(cm, classes,
normalize=False,
title='Confusion matrix',
cmap=plt.cm.Blues):
"""Plots the confusion matrix."""
if normalize:
cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
print("Normalized confusion matrix")
else:
print('Confusion matrix, without normalization')
plt.imshow(cm, interpolation='nearest', cmap=cmap)
plt.title(title)
plt.colorbar()
tick_marks = np.arange(len(classes))
plt.xticks(tick_marks, classes, rotation=55)
plt.yticks(tick_marks, classes)
fmt = '.2f' if normalize else 'd'
thresh = cm.max() / 2.
for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
plt.text(j, i, format(cm[i, j], fmt),
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black")
plt.ylabel('True label')
plt.xlabel('Predicted label')
plt.tight_layout()
# Classify pose in the TEST dataset using the trained model
y_pred = model.predict(X_test)
# Convert the prediction result to class name
y_pred_label = [class_names[i] for i in np.argmax(y_pred, axis=1)]
y_true_label = [class_names[i] for i in np.argmax(y_test, axis=1)]
# Plot the confusion matrix
cm = confusion_matrix(np.argmax(y_test, axis=1), np.argmax(y_pred, axis=1))
plot_confusion_matrix(cm,
class_names,
title ='Confusion Matrix of Pose Classification Model')
# Print the classification report
print('\nClassification Report:\n', classification_report(y_true_label,
y_pred_label))
Confusion matrix, without normalization
Classification Report:
precision recall f1-score support
chair 1.00 1.00 1.00 84
cobra 0.99 1.00 0.99 93
dog 1.00 1.00 1.00 84
tree 1.00 1.00 1.00 96
warrior 1.00 0.99 0.99 68
accuracy 1.00 425
macro avg 1.00 1.00 1.00 425
weighted avg 1.00 1.00 1.00 425
(Facultatif) Enquêter sur les prédictions incorrectes
Vous pouvez regarder les poses du TEST ensemble de données qui ont été incorrectement prédit pour voir si la précision du modèle peut être amélioré.
if is_skip_step_1:
raise RuntimeError('You must have run step 1 to run this cell.')
# If step 1 was skipped, skip this step.
IMAGE_PER_ROW = 3
MAX_NO_OF_IMAGE_TO_PLOT = 30
# Extract the list of incorrectly predicted poses
false_predict = [id_in_df for id_in_df in range(len(y_test)) \
if y_pred_label[id_in_df] != y_true_label[id_in_df]]
if len(false_predict) > MAX_NO_OF_IMAGE_TO_PLOT:
false_predict = false_predict[:MAX_NO_OF_IMAGE_TO_PLOT]
# Plot the incorrectly predicted images
row_count = len(false_predict) // IMAGE_PER_ROW + 1
fig = plt.figure(figsize=(10 * IMAGE_PER_ROW, 10 * row_count))
for i, id_in_df in enumerate(false_predict):
ax = fig.add_subplot(row_count, IMAGE_PER_ROW, i + 1)
image_path = os.path.join(images_out_test_folder,
df_test.iloc[id_in_df]['file_name'])
image = cv2.imread(image_path)
plt.title("Predict: %s; Actual: %s"
% (y_pred_label[id_in_df], y_true_label[id_in_df]))
plt.imshow(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
plt.show()
Partie 3 : convertir le modèle de classification des poses en TensorFlow Lite
Vous allez convertir le modèle de classification de pose Keras au format TensorFlow Lite afin de pouvoir le déployer sur des applications mobiles, des navigateurs Web et des appareils IoT. Lors de la conversion du modèle, vous appliquez la quantification de la plage dynamique pour réduire la classification de pose tensorflow Lite taille du modèle d'environ 4 fois avec une perte de précision insignifiante.
converter = tf.lite.TFLiteConverter.from_keras_model(model)
converter.optimizations = [tf.lite.Optimize.DEFAULT]
tflite_model = converter.convert()
print('Model size: %dKB' % (len(tflite_model) / 1024))
with open('pose_classifier.tflite', 'wb') as f:
f.write(tflite_model)
2021-12-21 12:12:00.560331: W tensorflow/python/util/util.cc:368] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them. INFO:tensorflow:Assets written to: /tmp/tmpr1ewa_xj/assets 2021-12-21 12:12:02.324896: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:363] Ignored output_format. 2021-12-21 12:12:02.324941: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:366] Ignored drop_control_dependency. WARNING:absl:Buffer deduplication procedure will be skipped when flatbuffer library is not properly loaded Model size: 26KB
Ensuite, vous écrivez le fichier d'étiquettes qui contient le mappage des index de classe vers les noms de classe lisibles par l'homme.
with open('pose_labels.txt', 'w') as f:
f.write('\n'.join(class_names))
Comme vous avez appliqué la quantification pour réduire la taille du modèle, évaluons le modèle TFLite quantifié pour vérifier si la baisse de précision est acceptable.
def evaluate_model(interpreter, X, y_true):
"""Evaluates the given TFLite model and return its accuracy."""
input_index = interpreter.get_input_details()[0]["index"]
output_index = interpreter.get_output_details()[0]["index"]
# Run predictions on all given poses.
y_pred = []
for i in range(len(y_true)):
# Pre-processing: add batch dimension and convert to float32 to match with
# the model's input data format.
test_image = X[i: i + 1].astype('float32')
interpreter.set_tensor(input_index, test_image)
# Run inference.
interpreter.invoke()
# Post-processing: remove batch dimension and find the class with highest
# probability.
output = interpreter.tensor(output_index)
predicted_label = np.argmax(output()[0])
y_pred.append(predicted_label)
# Compare prediction results with ground truth labels to calculate accuracy.
y_pred = keras.utils.to_categorical(y_pred)
return accuracy_score(y_true, y_pred)
# Evaluate the accuracy of the converted TFLite model
classifier_interpreter = tf.lite.Interpreter(model_content=tflite_model)
classifier_interpreter.allocate_tensors()
print('Accuracy of TFLite model: %s' %
evaluate_model(classifier_interpreter, X_test, y_test))
Accuracy of TFLite model: 1.0
Maintenant , vous pouvez télécharger le modèle TFLite ( pose_classifier.tflite ) et le fichier d'étiquette ( pose_labels.txt ) pour classer les poses personnalisées. Voir l' Android et Python / Raspberry Pi exemple d' application pour un exemple de bout en bout de la façon d'utiliser le modèle de classification de pose de TFLite.
zip pose_classifier.zip pose_labels.txt pose_classifier.tflite
adding: pose_labels.txt (stored 0%) adding: pose_classifier.tflite (deflated 35%)
# Download the zip archive if running on Colab.
try:
from google.colab import files
files.download('pose_classifier.zip')
except:
pass