Home

1. Move data to Sherlock

• rsync command
• test: size on server == size on sherlock, maybe frames (?)

2. Extract frame times / timestamps

• use show_frames to extract all frames and their timings (data/raw)
• parse text files into csv with frame + timing (inside data/parsed)
• (tests: potentially check with ffprobe)

3. Extract and Rotate frames

• ffmpeg commands (currently bash script calls python -- maybe consolidate?)
• test: ffprobe counts == # of images in output (create tests folder)
• data folder with output (this will probably have to be specified with link to $PI_SCRATCH, too big a folder)
• also pull a frame and check if needs to be rotated

4. Run MTCNN

• mtcnn library, bash script to run, specify output, etc
• test: # rows in CSV correspond to # frames

5. Run OpenPose on videos

• singularity / docker scripts? and bash script to use docker
• JSON -> CSV
• test: # rows == # frames
• data folder

6. Ground truth

• scripts to get ground truth
• data folder

7. Merge data?

• combine into giant csv
• other sanity checks
• all data goes here?

8. Preliminary Analyses

• rmd
• use data in step 6

9. Writeup

• use data in step 6