Open Hoantrbl opened 3 weeks ago
When I try to retrain the mdoel based on your checkpoint and save the better model pickle, I find that I can load my pickle successfully. I think it might be your checkpoint wrong.
Can you upload the lastest checkpoint?
Hello, this error comes from the config file, where the learnable ['MODEL']['TextEncoder']['leranable_ctx'] is needed to be True. I have updated the config file, please try to load the checkpoint again.
Thanks a lot!
In the meantime, I find that some errors occurs in your paper experiments.
In EventCLIP, it utilizes the VIT-B/16 for fine-tuning experiments.
However, in your paper, you said it's VIT-L/14, which may causes the unfair comparision and wrong analyisis in Model parameter size analysis.
I strongly recommend the author should make more clear explanation about it.
Please refer to the latest version of EventCLIP, page 5, Our Implementation Details, where they said 'For the pre-trained CLIP, we adopt the variant with the ViT-L/14 [10] image encoder'.
Please refer to the latest version of EventCLIP, page 5, Our Implementation Details, where they said 'For the pre-trained CLIP, we adopt the variant with the ViT-L/14 [10] image encoder'.
You can refer to https://github.com/Wuziyi616/EventCLIP/issues/5#issuecomment-2306389902 for EventCLIP's response.
When I try to load the checkpoint about the N-Caltech101, it occurs the following error:
It seems like something (maybe the "SpecificTextualPrompt" module) can't not be loaded into the model. Then I try to correct the error by utilizing the following code:
The errors disappear. However, I wonder whether it will influence the performance of the model?
Can you help me fix the code?