Video Action Classification using 3D ResNet

Reference:

https://github.com/kenshohara/video-classification-3d-cnn-pytorch

(Linux)

1. Install Anaconda and Setup Pytorch (GPU)

2. Download code from Github (https://github.com/kenshohara/video-classification-3d-cnn-pytorch)

3. Download Pre-trained Model and copy to source code folder (resnext-101-kinetics.pth)

   https://drive.google.com/drive/folders/1zvl89AgFAApbH0At-gMuZSeQB_LpNP-M

4. Install FFmpeg

   conda install -c anaconda ffmpeg

5. Run a code

python main.py --input ./input --video_root ./videos --output ./output.json --model ./resnext-101-kinetics.pth --resnet_shortcut B  --mode score --model_name resnext --model_depth 101

6. You can get result file (output.json) "segment": [start frame number, end frame number], "label": " "

7. Generate labeled video (Option)

python generate_result_video/generate_result_video.py ../output.json ../videos ./videos_pred ../class_names_list 5

(Windows) -- Not Recommend

1. Install Anaconda and Setup Pytorch (GPU)

2. Download code from Github (https://github.com/kenshohara/video-classification-3d-cnn-pytorch)

3. Download Pre-trained Model and copy to source code folder (resnext-101-kinetics.pth)

   https://drive.google.com/drive/folders/1zvl89AgFAApbH0At-gMuZSeQB_LpNP-M

4. Download FFmpeg and copy to source code folder

5. Change code (subprocess.call() is for removing and creating folders in linux)

(shutil is need for windows) import shutil subprocess.call('rm -rf tmp', shell=True)  --> (Change like this)   if os.path.exists('tmp'):      shutil.rmtree('tmp') subprocess.call('mkdir tmp', shell=True)  --> (Change like this) os.makedirs('tmp')

6. Run a code

python main.py --input ./input --video_root ./videos --output ./output.json --model ./resnext-101-kinetics.pth --resnet_shortcut B  --mode score --model_name resnext --model_depth 101

7. You can get result file (output.json) "segment": [start frame number, end frame number], "label": " "

8. Generate labeled video (Option) (It's not easy to change the code to run on Windows) 

   --> You'd better to run a code on Linux