search

594422814 / UDT_pytorch

MIT License
191 stars 23 forks source link

Where cost-sensitive loss is computed? #5

Closed ghost closed 5 years ago

ghost commented 5 years ago

Hi. I read your paper and got very interested in your unsupervised tracking.

My question is that where in the codes cost-sensitive loss (A^i_{motion}) is implemented. I found train_UDT.py includes reduction of samples with very high loss values (A^i_{drop}), but I cannot find where it weights samples according to the motion. Is it not implemented yet?

594422814 commented 5 years ago

Sorry, I did not implement this. The results reported in the paper are obtained using the Matlab code. The I wrote this pytorch version. Since the A_i_drop is necessary (otherwise we observe that our unsupervised loss cannot well converge). But I forgot to implement the cost-sensitive loss. I will add this later.

ghost commented 5 years ago

I see. Thank you!

swrdZWJ commented 5 years ago

I meet the same problem and when do you release the lastest and complete version with implementation of cost-sensitive loss？Thanks～～

ReedZyd commented 4 years ago

When do you realse the latest and completed version with implementation of cost-sensitive loss?