search

pawn-sxy / DCAMA

This is the official implementation of the ECCV'2022 paper "Dense Cross-Query-and-Support Attention Weighted Mask Aggregation for Few-Shot Segmentation".
61 stars 15 forks source link

About 5-shot training #1

Closed Strike1999 closed 2 years ago

Strike1999 commented 2 years ago

Hello !

Thanks for your exciting job ! I can test the model in 5-shot setting , but how can i train the model in 5-shot . It seems like this commit only provides codes for 1-shot training .

Thanks a lot .

pawn-sxy commented 2 years ago

We reuse the 1-shot trained model for k-shot inference for computational efficiency, but it is also possible to train the model under 5-shot setting.

If you want to do so, please refer to the "predict_mask_nshot()" method and change the calculation of logit_mask in "forward()" method, both of them are in DCAMA.py.

Strike1999 commented 2 years ago

I know . Thanks for reply !