Can smaller pre-trained Model be used to train?

tianrun-chen / SAM-Adapter-PyTorch

Adapting Meta AI's Segment Anything to Downstream Tasks with Adapters and Prompts

MIT License

968 stars 83 forks source link

Can smaller pre-trained Model be used to train? #13

Closed hhhhyyyyyyrrrrr closed 1 year ago

tianrun-chen commented 1 year ago

Hello! We have updated the config file, so that you can choose to use the ViT-H, ViT-B, or ViT-L version of SAM. Of course, you can also use different backbone structures to replace the SAM.

hhhhyyyyyyrrrrr commented 1 year ago

Ok, thank you very much for your reply！

Heroman2Space commented 9 months ago

But it seems we cannot simply use the original ViT-H, ViT-B, or ViT-L version of SAM. There is a lot of mismatched keys when loading the params.