search

Tsingularity / FRN

(CVPR 2021) Few-Shot Classification with Feature Map Reconstruction Networks
MIT License
110 stars 18 forks source link

Performance on CUB #5

Closed LoveMiki closed 2 years ago

LoveMiki commented 2 years ago

Thanks for sharing the codes and your excellent work.

In Table 3, I notice that ProtoNet achieves 81.50 with pre-training and 83.88 under your implementation using the basic Conv-4 backbone. Then, in 9.1, I find that you said "Conv-4 are trained with 20-way 5-shot for 5 shot model, and 30-way 1-shot for 1-shot model". So, the accuracy of 83.88 under your implementation is trained from scratch using the above settings with SGD? Did you implement ProtoNet using the general setting (e.g., 5-way 5-shot and 1-shot)?

Sorry to bother you and thank you in advance.

Tsingularity commented 2 years ago

Yes, in the main CUB table, we don't do pre-training for our own implemented ProtoNet model. And you should be able to reproduce the numbers using our released code.

I didn't train with 5-way 5/1-shot episodes on this particular codebase, but feel free to give it a try! And also feel free to let us know if there's anything we can help with.

Tsingularity commented 2 years ago

closing this for now. feel free to re-open it if you need more help.