usage.txt

Fullbody Single Person Filtering from YouTube Videos(fullbody_filter.py)

This will download videos from YouTube and save images and masks that contain single full body person by using OpenPose keypoints to determine whether the face, hands, and feet are in the given frame and Mask-RCNN to generate the person mask.
*Requires generate_mask.py and downloadVideo.py and Pytube 
*Pytube downloading videos sometimes does not work because Google changes Youtube internally. If this is the case, then set skip_download=True and manually download until a patch is ready.

How to use:
 - set all parameters(save paths, youtube links,...), line 101
IMPORTANT: The video_save_path and op_save_path folders will be emptied prior to
processing
- op_save_path is temporary folder to store initial images filtered by OpenPose that will be accessed by Mask-RCNN in second half of script
- in generate_masks.py, make sure ROOT_DIR is set to mask_rcnn folder line 12
- To run: python3 fullbody_filter.py 
- saved images will be of form, [name of video]_[img number].jpg
  

ME vs RCNN comparison(me_rcnn_compare.py)

Compares masks generated by Mask-RCNN and ME by calculating IOU and saving those under a certain threshold. Can also generate side by side comparison of ME vs Mask-RCNN 
*Can run anywhere, no dependencies

How to use:
- set all parameters(save_paths, threshold,...) line 61
- side_by_side variable controls whether to create side by side comparison
- To run: python3 me_rcnn_compare.py