search

syp2ysy / SVF

[NeurIPS 2022] Singular Value Fine-tuning: Few-shot Segmentation requires Few-parameters Fine-tuning
MIT License
69 stars 6 forks source link

Question about baseline #1

Closed TinyAway closed 2 years ago

TinyAway commented 2 years ago

Thanks for sharing your interesting work. I have a question when studying your paper. It is observed that you got better results using ASPP instead of FEM in PFENet as shown in Table 12. However, this phenomenon is opposite to the result in the original paper. Did you use any extra tricks to get these results?

syp2ysy commented 2 years ago

@TinyAway Thanks for your interest. Some details in our code are different from original pfenet code, for example, the batchnorm in the backbone needs to be adjusted to the .eve() mode; using DistributedDataParallel with SyncBatchNorm. In the same framework, pfenet performs much better than ASPP, fairer results are in table1 and table2. Following previous paper, the results of comparison methods in Table 12 were obtained directly from the original paper.