output

Output

Each PNG should be named according to the hash of the video it predicts segmentations for, and should have a “2” for every pixel predicted to be cilia any other non-2 value is fine for other pixels, since they’ll be ignored anyway.Once your model has made its predictions and created the image files of the masks, tar them all together