microsoft / RegionCLIP

[CVPR 2022] Official code for "RegionCLIP: Region-based Language-Image Pretraining"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

AttributeError: 'ModifiedResNet' object has no attribute 'num_features'

gjc0824 opened this issue · comments

I am sorry to bother you. I run your new zero-shot branch but I receive some bugs. I check and find out that this part doesn't exisit in main branch. Could you please help me solve it?
image

thanks for pointing this out! I will take a look and get back to you asap.

Thanks for your reply! I should provide you more infomation about this bug. It occurs when I use the config file of your original resnet version. It won't occur when I use the vit version config file. Maybe the two branch codes are not suitable for each other.
And I also have a naive question. There are only config files that support the other versions such as vit and swin-t in this project. However, I don't find the checkpoint or weights file support other versions. So is there pretrained RegionCLIP that using transformer-version backbone? Maybe I misunderstand the model.
Looking forward your reply! Thanks for your exciting contribution!

thanks for the context! still looking at it

Besides resnet, I couldn't find a pre trained model either

Hi, have you solved this problem? I also encountered the same problem. Your help is greatly needed!

Thanks for your reply! I should provide you more infomation about this bug. It occurs when I use the config file of your original resnet version. It won't occur when I use the vit version config file. Maybe the two branch codes are not suitable for each other. And I also have a naive question. There are only config files that support the other versions such as vit and swin-t in this project. However, I don't find the checkpoint or weights file support other versions. So is there pretrained RegionCLIP that using transformer-version backbone? Maybe I misunderstand the model. Looking forward your reply! Thanks for your exciting contribution!

@gjc0824 Were you able to train the ViT version?