Module deepmind/i3d-kinetics-600/1

Inflated 3D Convnet or I3D model [1] trained for action recognition on kinetics-600.

Module URL:

A Colaboratory notebook demonstrating a simple application of this module is available here.


This video classification model is described in [1], the source code is publicly available on github.

As reported in [1], this model achieved state-of-the-art results on the UCF101 and HMDB51 datasets from fine-tuning these models. I3D models pre-trained on Kinetics also placed first in the CVPR 2017 Charades challenge.

Example use

frames = ... # Shape [batch_size, frame_count, height=224, width=224, 3]
module = hub.Module("")
logits = module(frames)

The labels for the 600 different actions are detailed in this map.


[1] Joao Carreira and Andrew Zisserman. Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset. Conference on Computer Vision and Pattern Recognition, CVPR 2017.