submission name:

mask2mask

upload zipped archive.zip

archive.zip

report.txt

=== Evaluation config ===
  batch_size: 16
  device: cuda
  num_threads: 2
  prefetch_queue_depth: 4
  report: submissions/mask2mask/report.txt
  seed: 1234
  submission_dir: submissions/mask2mask
  uncompressed_dir: /mnt/d/comma_submission/comma_video_compression_challenge/videos
  video_names_file: /mnt/d/comma_submission/comma_video_compression_challenge/public_test_video_names.txt
=== Evaluation results over 600 samples ===
  Average PoseNet Distortion: 0.00065614
  Average SegNet Distortion: 0.00264182
  Submission file size: 386,192 bytes
  Original uncompressed size: 37,545,489 bytes
  Compression Rate: 0.01028598
  Final score: 100*segnet_dist + √(10*posenet_dist) + 25*rate = 0.60

does your submission require gpu for evaluation (inflation)?

yes

did you include the compression script? and want it to be merged?

no

additional comments

In theory it should still run on CPU in <30 min. The submission name is kind of true but doesn't really encompass what the model does. At least its not a ffmpeg settings tweak.

On an unrelated note, I think the problem with challenges that also have open source submissions is that people are incentivized to not submit until the last minute to prevent other people from tweaking some tiny thing and surpassing their score. I kind of wanted to see some cool submissions, but I'm assuming the reason there's none yet is because they're saving all of their competitive submissions. I think <0.50 is pretty easily possible as a slightly different architecture gets a score at least 10% better, so there's definitely a lot of room to go. I minimally obfuscated the exact model architecture so people can't trivially spend more compute and clog up the leaderboards with exact clones to try to get prizes.