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

. #5

Closed JawadTawhidi closed 1 year ago

JawadTawhidi commented 1 year ago

.

suhwan-cho commented 1 year ago

Hi Jawad,

In VOS field, the word "unsupervised" is usually equal to "zero-shot". It means that there is no guidance on the object to segment during inference. In this case, "unsupervised" has nothing to do with network training protocol. During testing, TMO does not use ground truth masks for inference. They are just loaded for performance evaluation.

JawadTawhidi commented 1 year ago

Thank you so much for your response.