Our paper "Various frameworks for integrating image and video streams for spatiotemporal information learning employing 2D-3D Residual networks for human action recognition" is under review
This is the PyTorch code for the video-based recognition
resnet-101-kinetics.pth: --model resnet --model_depth 101 --resnet_shortcut B
** that fine-tuned models on UCF-101 and HMDB-51
* Download videos and train/test splits [here](http://crcv.ucf.edu/data/UCF101.php).
* Convert from avi to jpg files using ```utils/video_jpg_ucf101_hmdb51.py```
##bash code
python utils/video_jpg_ucf101_hmdb51.py avi_video_directory jpg_video_directory
* Generate n_frames files using ```utils/n_frames_ucf101_hmdb51.py```
##bash code
python utils/n_frames_ucf101_hmdb51.py jpg_video_directory
- Generate annotation file in json format similar to ActivityNet using
utils/ucf101_json.pyannotation_dir_pathincludes classInd.txt, trainlist0{1, 2, 3}.txt, testlist0{1, 2, 3}.txt
python utils/ucf101_json.py annotation_dir_path* Download videos and train/test splits [here](http://serre-lab.clps.brown.edu/resource/hmdb-a-large-human-motion-database/).
* Convert from avi to jpg files using ```utils/video_jpg_ucf101_hmdb51.py```
python utils/video_jpg_ucf101_hmdb51.py avi_video_directory jpg_video_directory- Generate n_frames files using
utils/n_frames_ucf101_hmdb51.py
python utils/n_frames_ucf101_hmdb51.py jpg_video_directory
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent