Mianzhi Pan

I use the extracted RoI image features same as [ROSITA](https://github.com/MILVLG/rosita). And I directly concat the 2048-d image features and the 6-D box features (x_min, y_min, x_max, y_max, height, width) as...

I follow the demo and use the features generated by `albef_feature_extraction` to perform zero-shot cross-modal retrieval on MSCOCO. The t2i recall scores are extremely low while that of i2t score...

Can not reproduce the result on DrugOOD.

1

I ran your code on DrugOOD with 10 different seeds (0 ~ 9), but the results are significantly lower than those reported in your paper. For example, the ROC-AUC on...

The following is quoted from the paper: > First, we test the robustness of experimental performances to the change of feature scale. Specifically, we use L2 regularization to change the...