इमेजनेट2012_मल्टीलेबल

  • विवरण :

इस डेटासेट में ILSVRC-2012 (इमेजनेट) सत्यापन छवियां शामिल हैं, जिन्हें "इमेजनेट पर मूल्यांकन मशीन सटीकता" , ICML, 2020 से मल्टी-क्लास लेबल के साथ एनोटेट किया गया है। ImageNet वर्ग पदानुक्रम में दानेदार वर्ग भेद (अधिक विवरण के लिए पेपर देखें)। मूल लेबल की तुलना में, ये विशेषज्ञ-समीक्षित बहु-श्रेणी के लेबल सटीकता के अधिक अर्थपूर्ण रूप से सुसंगत मूल्यांकन को सक्षम करते हैं।

इस डेटासेट के संस्करण 3.0.0 में "आटा कब बैगेल बनता है? इमेजनेट पर शेष गलतियों का विश्लेषण करने के साथ-साथ इमेजनेट-मेजर (इमेजनेट-एम) 68-उदाहरण 'इमेजनेट-एम' के तहत विभाजन से अधिक सही लेबल शामिल हैं।

50,000 इमेजनेट सत्यापन छवियों में से केवल 20,000 में मल्टी-लेबल एनोटेशन हैं। मल्टी-लेबल का सेट पहले 67 प्रशिक्षित इमेजनेट मॉडल के टेस्टबेड द्वारा उत्पन्न किया गया था, और फिर प्रत्येक व्यक्तिगत मॉडल की भविष्यवाणी को विशेषज्ञों द्वारा मैन्युअल रूप से एनोटेट किया गया था (छवि के लिए लेबल correct है), wrong (लेबल छवि के लिए गलत है) छवि), या unclear (विशेषज्ञों के बीच कोई आम सहमति नहीं बनी)।

इसके अतिरिक्त, एनोटेशन के दौरान, विशेषज्ञ पैनल ने समस्याग्रस्त छवियों के एक सेट की पहचान की। एक छवि समस्याग्रस्त थी यदि वह नीचे दिए गए किसी भी मानदंड को पूरा करती थी:

  • मूल इमेजनेट लेबल (टॉप-1 लेबल) गलत या अस्पष्ट था
  • छवि एक ड्राइंग, पेंटिंग, स्केच, कार्टून, या कंप्यूटर-प्रदत्त थी
  • छवि अत्यधिक संपादित की गई थी
  • चित्र में अनुपयुक्त सामग्री थी

समस्याग्रस्त छवियां इस डेटासेट में शामिल हैं लेकिन बहु-लेबल सटीकता की गणना करते समय इसे अनदेखा किया जाना चाहिए। इसके अतिरिक्त, चूंकि 20,000 एनोटेशन का प्रारंभिक सेट वर्ग-संतुलित है, लेकिन समस्याग्रस्त छवियों का सेट नहीं है, हम प्रति-वर्ग सटीकता की गणना करने और फिर उनका औसत करने की सलाह देते हैं। हम भविष्यवाणी को सही मानने की भी सलाह देते हैं यदि इसे सही या अस्पष्ट के रूप में चिह्नित किया गया है (अर्थात, अस्पष्ट लेबल के साथ उदार होना)।

ऐसा करने का एक संभावित तरीका निम्न NumPy कोड के साथ है:

import tensorflow_datasets as tfds

ds = tfds.load('imagenet2012_multilabel', split='validation')

# We assume that predictions is a dictionary from file_name to a class index between 0 and 999

num_correct_per_class = {}
num_images_per_class = {}

for example in ds:
    # We ignore all problematic images
    if example[‘is_problematic’].numpy():
        continue

    # The label of the image in ImageNet
    cur_class = example['original_label'].numpy()

    # If we haven't processed this class yet, set the counters to 0
    if cur_class not in num_correct_per_class:
        num_correct_per_class[cur_class] = 0
        assert cur_class not in num_images_per_class
        num_images_per_class[cur_class] = 0

    num_images_per_class[cur_class] += 1

    # Get the predictions for this image
    cur_pred = predictions[example['file_name'].numpy()]

    # We count a prediction as correct if it is marked as correct or unclear
    # (i.e., we are lenient with the unclear labels)
    if cur_pred is in example['correct_multi_labels'].numpy() or cur_pred is in example['unclear_multi_labels'].numpy():
        num_correct_per_class[cur_class] += 1

# Check that we have collected accuracy data for each of the 1,000 classes
num_classes = 1000
assert len(num_correct_per_class) == num_classes
assert len(num_images_per_class) == num_classes

# Compute the per-class accuracies and then average them
final_avg = 0
for cid in range(num_classes):
  assert cid in num_correct_per_class
  assert cid in num_images_per_class
  final_avg += num_correct_per_class[cid] / num_images_per_class[cid]
final_avg /= num_classes

  • होमपेज : https://github.com/modestyachts/evaluating_machine_accuracy_on_imagenet

  • स्रोत कोड : tfds.datasets.imagenet2012_multilabel.Builder

  • संस्करण :

    • 1.0.0 : प्रारंभिक रिलीज।
    • 2.0.0 : निश्चित ILSVRC2012_img_val.tar फ़ाइल।
    • 3.0.0 (डिफ़ॉल्ट): संशोधित लेबल और ImageNet-M विभाजन।
  • डाउनलोड आकार : 191.13 MiB

  • डेटासेट का आकार : 2.50 GiB

  • मैन्युअल डाउनलोड निर्देश : इस डेटासेट के लिए आपको स्रोत डेटा को मैन्युअल रूप से download_config.manual_dir (डिफ़ॉल्ट रूप से ~/tensorflow_datasets/downloads/manual/ ) में डाउनलोड करना होगा:
    मैन्युअल_डीआईआर में ILSVRC2012_img_val.tar फ़ाइल होनी चाहिए। डेटासेट डाउनलोड करने के लिए लिंक प्राप्त करने के लिए आपको http://www.image-net.org/download-images पर पंजीकरण करना होगा।

  • ऑटो-कैश्ड ( दस्तावेज़ीकरण ): नहीं

  • विभाजन :

विभाजित करना उदाहरण
'imagenet_m' 68
'validation' 20,000
  • फ़ीचर संरचना :
FeaturesDict({
    'correct_multi_labels': Sequence(ClassLabel(shape=(), dtype=int64, num_classes=1000)),
    'file_name': Text(shape=(), dtype=string),
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'is_problematic': bool,
    'original_label': ClassLabel(shape=(), dtype=int64, num_classes=1000),
    'unclear_multi_labels': Sequence(ClassLabel(shape=(), dtype=int64, num_classes=1000)),
    'wrong_multi_labels': Sequence(ClassLabel(shape=(), dtype=int64, num_classes=1000)),
})
  • फ़ीचर दस्तावेज़ीकरण :
विशेषता कक्षा आकार डीटाइप विवरण
विशेषताएं डिक्ट
सही_बहु_लेबल अनुक्रम (कक्षा लेबल) (कोई भी नहीं,) int64
फ़ाइल का नाम मूलपाठ डोरी
छवि छवि (कोई नहीं, कोई नहीं, 3) uint8
is_problematic टेन्सर बूल
original_लेबल क्लासलेबल int64
अस्पष्ट_बहु_लेबल अनुक्रम (कक्षा लेबल) (कोई भी नहीं,) int64
गलत_बहु_लेबल अनुक्रम (कक्षा लेबल) (कोई भी नहीं,) int64

VISUALIZATION

  • उद्धरण :
@article{shankar2019evaluating,
  title={Evaluating Machine Accuracy on ImageNet},
  author={Vaishaal Shankar* and Rebecca Roelofs* and Horia Mania and Alex Fang and Benjamin Recht and Ludwig Schmidt},
  journal={ICML},
  year={2020},
  note={\url{http://proceedings.mlr.press/v119/shankar20c.html} }
}
@article{ImageNetChallenge,
  title={ {ImageNet} large scale visual recognition challenge},
  author={Olga Russakovsky and Jia Deng and Hao Su and Jonathan Krause
   and Sanjeev Satheesh and Sean Ma and Zhiheng Huang and Andrej Karpathy and Aditya Khosla and Michael Bernstein and
   Alexander C. Berg and Fei-Fei Li},
  journal={International Journal of Computer Vision},
  year={2015},
  note={\url{https://arxiv.org/abs/1409.0575} }
}
@inproceedings{ImageNet,
   author={Jia Deng and Wei Dong and Richard Socher and Li-Jia Li and Kai Li and Li Fei-Fei},
   booktitle={Conference on Computer Vision and Pattern Recognition (CVPR)},
   title={ {ImageNet}: A large-scale hierarchical image database},
   year={2009},
   note={\url{http://www.image-net.org/papers/imagenet_cvpr09.pdf} }
}
@article{vasudevan2022does,
  title={When does dough become a bagel? Analyzing the remaining mistakes on ImageNet},
  author={Vasudevan, Vijay and Caine, Benjamin and Gontijo-Lopes, Raphael and Fridovich-Keil, Sara and Roelofs, Rebecca},
  journal={arXiv preprint arXiv:2205.04596},
  year={2022}
}