CLIP4Clip

CLIP4Clip copied to clipboard

有没有朋友遇到这个问题，是怎么解决的 呢

Mornnn

Thank you for your great code. Actually, I found that the dataset sampler does not get any seed, so it use default value '0' as seed. It seems that the...

goonbamm

Results on MSRVTT and MSVD

3

Thanks for your great work! I successfully reproduced the results of meanP on MSRVTT and MSVD, where I exactly followed your configuration on readme. But for the experiments of seqTransf,...

STARK1234

Use my own videos

2

I want to perform video retrieval using my own videos. Can you tell me how to use my own videos for performing video-language retrieval?

akshaymanj

Could some guys share codes, please?

fishfree

Large performance drop if trained with fp32.

1

Hi authors, Thanks for your great work! In the file [module_clip](https://github.com/ArrowLuo/CLIP4Clip/tree/master/modules/module_clip.py) @ L557: ``` python convert_weights(model) model.load_state_dict(state_dict) return model.eval() ``` If I remove `convert_weight`, the model can only achieve an...

klauscc

train files

1

Hello, can you share "train.json" and "val_1.json" of activatitynet dataset.?

1617226214

It is very unfriendly to directly pass the entire config as a argument to a function, e.g., https://github.com/ArrowLuo/CLIP4Clip/blob/508ffa3de39ba0563a03199c440ab602a72e9b6f/modules/modeling.py#L30

deepalchemist

Thanks to your code, I am growing every day. Thank you very much. ---- In every dataloader, the special tokens are initialized below. ~~~python self.SPECIAL_TOKEN = {"CLS_TOKEN": "", "SEP_TOKEN": "",...

goonbamm

Dear Author, I feel always thankful for your reply. Q1. Here is another quetion, In the training code, is there any difference in performance if I change the batch_size_eval 16...

celestialxevermore