search

grantword8 / BLV

Balancing Logit Variation for Long-tailed Semantic Segmentation
21 stars 0 forks source link

can't reproduce result #6

Open seonggyuny opened 11 months ago

seonggyuny commented 11 months ago

Hi, Thanks for your great work! I have a question about the code. In blv_loss.py line19 : self.frequency_list = torch.log(sum(cls_num_list)) - frequency_list the line have error because the cls_num_list is not tensor, so the code Shouldn't the code be modified like this? "self.frequency_list = torch.log(sum(cls_list)) - frequency_list" and I can't reproduce your results about mIoU. mIoU using the config file is different from your work. can you check this? my env is pytorch=1.9.1, cuda=11.2 and others is same with you.

Thank you.

dolphin0104 commented 11 months ago

Same question! I also have additional question.. I also wonder if the provided 'cls_num_list' distribution (below) in the code is measured using Cityscape training dataset. cls_num_list=[16099193283, 4155428633, 8478285320, 925367586, 317481282, 531401266, 67288808, 40463971, 3810296251, 1074861777, 6784666433, 181828023, 15321782, 259269812, 564950248, 184638969, 32522117, 15799033, 2718199]

userid623 commented 7 months ago

In fact, this paper is very similar to a CVPR '22 paper GCL. I am even inclined to consider it a case of plagiarism. Given your inability to reproduce the results, I am now highly skeptical that the experiments were fabricated.

GCL: Long-tailed visual recognition via gaussian clouded logit adjustment