mala-lab / sic-cads Goto Github PK

Hi, thanks for the great work!
I'm trying to train the model with online training strategy but cannot proceed with the provided code.
I tried changing the online_train configuration to True but think there's more modifications need to be applied on the code.
Could you provide the exact code and config files for the online training strategy?
Thanks again for the interesting work.

I want to know where you get the file resource/coco_there_is_a_cls_vitb32.pt . And if you create it by yourself , please tell me how to do that. Thanks a lot!

Can you provide the code for directly applying the Multi-modal MLR trained in SIC-CADS to CORA?

I tried to retrain the MLR moudle and not found the file datasets/coco/zero-shot/instances_train2017_seen_2_oriorder_cat_info.json

Do I miss something needed? Thanks.

Thanks for this interesting work.

This paper uses cos_sim to compute the simliarity between Learned Text Embeddings and CLIP Text Embeddings,But I can find out where it's using it.

if not self.multi_scale:
            pred_ml_scores = self.logit_scale * self.text_embedding(text_features)
        else:
            pred_ml_scores = self.logit_scale * self.get_multi_level_scores(text_features)
        
        mlr_loss = self.get_rank_loss(pred_ml_scores, batched_inputs)

There doesn't seem to be a calculation going on here.

Can you provide BoxSup-C2_Lbase_CLIP_R5021k_640b64_4x_mlr.pth?
Thanks!

Thanks for this interesting work.

This paper uses a normalized score with the sigmoid function to calculate the MLR score.

Why not use the softmax to obtain the MLR score?

Can you provide BoxSup-C2_Lbase_CLIP_R5021k_640b64_4x_mlr.pth，Detic_LbaseI_CLIP_R5021k_640b64_4x_ft4x_max-size_mlr.pth and Detic_OVCOCO_CLIP_R50_1x_max-size_caption_mlr.pth for us to evaluate directly?