Where can we find the weights of the pretrained CLIP:ResNet-50 ?

[saimeur]

Hello,

First of all, thank you for your excellent work—both the paper and this repository are incredibly insightful.

I noticed in the CNNDetection repository that they achieved better results using uncropped images. I’m interested in experimenting with this approach using your model. However, since CLIP:ViT models are sensitive to image dimensions, I’m considering using CLIP:ResNet models instead (especially CLIP:RN50 as mentionned in your paper). Although CLIP:ResNet models tend to yield slightly lower performance compared to CLIP:ViT, they offer greater flexibility in handling varying image dimensions, which could be beneficial for this experiment.

Do you still have the pretained weights of CLIP:RN50 ?

[jim-xie-cn]

Hello,

I have same requirement as Saimeur, I want to compare the performance of models based on the different backbone network (such as CLIP:RN50,Imagenet:vgg19,...etc. ).

Do you still have the pretained weights which you mentioned in the paper ?

Thanks so much!