search

suhwan-cho / TMO

[WACV 2023] Treating Motion as Option to Reduce Motion Dependency in Unsupervised Video Object Segmentation
MIT License
56 stars 3 forks source link

. #7

Closed JawadTawhidi closed 1 year ago

JawadTawhidi commented 1 year ago

.

suhwan-cho commented 1 year ago

Hi, sorry for the late reply.

  1. You can use either test_davis, test_fbms, or test_ytobj for evaluation on new dataset. I have no idea about how the ultrasound videos are constructed, but formats would not be a matter if the annotations are binary.
  2. Yes, the evaluation code attached in TMO is identical to the official DAVIS evaluation code.
suhwan-cho commented 1 year ago
  1. If you want to use your own dataset instead of DAVIS, you have to make it have the same format as DAVIS. The background pixels should have the value of [0,0,0] and the foreground pixels should have the value of [128,0,0]. All pixels should have these values and no other values are accepted.
  2. In my device, I have obtained the same results. You can just ignore it as the difference is very small.
suhwan-cho commented 1 year ago

Yes, just make sure that the annotations have the pixel values of either [0,0,0] or [128,0,0]. I recommend you to check the pixel values and modify them if needed. I cannot tell if it works because I didn't see your data.

suhwan-cho commented 1 year ago

Hi, you can check the pixel values by converting PIL images to tensor format.

suhwan-cho commented 1 year ago

Yes, that's also correct.