The ViT-DD model uses images for training and testing. However, the DMD dataset contains videos. So, to run ViT-DD on the DMD dataset, we first need to extract individual frames from the DMD dataset. After some research, I have determined that FFmpeg is a good tool for this task. This tool is available at https://ffmpeg.org/. Please take a look at its documentation for your specific operating system to install it. This installation process is, however, not a trivial task to do. Take Windows as an example. You may need to unzip the downloaded file and add the folder to PATH. After installing FFmpeg, the workflow for this task is listed below:
Download, unzip, and explore the DMD dataset. For now, we are only interested in the RGB body and RGB face videos. Please identify these videos and delete all others to save your disk space.
Please create the following directories on your local machine: ./datasets/DMD/SideBody and ./datasets/DMD/FrontBody.
Please use FFmpeg to extract frames from the DMD videos you just downloaded and unzipped. When extracting frames, we only keep one frame every thirty frames. Please save the RGB body frames to ./datasets/DMD/SideBody and RGB face frames to ./datasets/DMD/FrontBody. When saving the frames, please follow the pattern specified using this pattern.
for the frames from the gA1 -> s1 -> RGB face, use the name ga1s1frontbody_xxxxx.png, where xxxxx is the frame number.
for the frames from the gA1 -> s1 -> RGB body, use the name ga1s1sidebody_xxxxx.png, where xxxxx is the frame number.
Modify the .gitignore to exclude the two directories created above so that the frames are not uploaded to GitHub due to space concerns.
Update the README.md with the information on how to use FFmpeg and where to store the extracted frames.
Use a USB drive or portable SSD to save and transport the frames between group members.
NOTE: Remember to take care of the front & side frame offsets when extracting the frames from both videos. Try to specify the frame-extracting software to take the 1st frame for front, and take the offsetted frame for the side.
Please finish the task as soon as possible on all the DMD videos provided by the TA. This project cannot proceed without these frames.
The ViT-DD model uses images for training and testing. However, the DMD dataset contains videos. So, to run ViT-DD on the DMD dataset, we first need to extract individual frames from the DMD dataset. After some research, I have determined that FFmpeg is a good tool for this task. This tool is available at https://ffmpeg.org/. Please take a look at its documentation for your specific operating system to install it. This installation process is, however, not a trivial task to do. Take Windows as an example. You may need to unzip the downloaded file and add the folder to PATH. After installing FFmpeg, the workflow for this task is listed below:
./datasets/DMD/SideBodyand./datasets/DMD/FrontBody../datasets/DMD/SideBodyand RGB face frames to./datasets/DMD/FrontBody. When saving the frames, please follow the pattern specified using this pattern.ga1s1frontbody_xxxxx.png, wherexxxxxis the frame number.ga1s1sidebody_xxxxx.png, wherexxxxxis the frame number.NOTE: Remember to take care of the front & side frame offsets when extracting the frames from both videos. Try to specify the frame-extracting software to take the 1st frame for front, and take the offsetted frame for the side.
Please finish the task as soon as possible on all the DMD videos provided by the TA. This project cannot proceed without these frames.