video action recognition

Video classification using Dataset: HockeyFights

c3d-pytorch
├── extract_frames.py
├── hockey_data.py
├── hockey_model.py
├── test.txt
├── train.txt
├── log.py
├── hockey_train.h5
├── hockey_test.h5
├── data
│   ├── fi100_xvid
│   ├── fi101_xvid
│   ├── fi102_xvid
│   │   ├── 0.jpg
│   │   ├── 1.jpg
│   │   ├── 2.jpg
│   │   ├──  ...
│   ├── ...
├── video
│   ├── HockeyFights
│   │   ├── fi100_xvid.avi
│   │   ├── fi101_xvid.avi
│   │   ├── fi102_xvid.avi
│   │   ├──  ...

Setup

Data Preparation

Please refer to data_preparation.md for a general knowledge of data preparation. The supported datasets are listed in supported_datasets.md.

We also share our Kinetics-400 annotation file k400_val, k400_train for better comparison.

Inference

Training

To train a video recognition model with pre-trained image models (for Kinetics-400 and Kineticc-600 datasets), run:

Colaboratory file: C3D_Hockey

Results and Models

HockeyFights

Model	Input size	acc
C3D	16 x 70 x 110	93.50
R21D	8 x 112 x 112	96.00
P3D	16 x 160 x 160	98.00
ARTNet	16x112x112	98.00