search

MILVLG / bottom-up-attention.pytorch

A PyTorch reimplementation of bottom-up-attention models
Apache License 2.0
294 stars 76 forks source link

Evaluation of attributes #89

Open drigoni opened 2 years ago

drigoni commented 2 years ago

Dear authors,

thanks for this excellent repository. I see in the evaluation code that some parts of the evaluation of the attributes are implemented, but I do not see an option to evaluate the attribute predictions. I tried here to change the flag to True, but some errors occur. Do I miss something?

Best regards, Davide

MIL-VLG commented 2 years ago

Currently, we do not support attribute prediction, as the original implementation in Caffe is not that straightforward to migrate to the d2 backend. They need another forward pass with the ground-truth class label as input for attribute prediction.

If you are interested, you can try implementing it yourself and a pull request is appreciated.