search

hitachinsk / SAMed

The implementation of the technical report: "Customized Segment Anything Model for Medical Image Segmentation"
MIT License
488 stars 46 forks source link

Position embedding mismatch. #28

Open zhixue-fang opened 1 year ago

zhixue-fang commented 1 year ago

The position embedding in the ViT-B weight file is for 1024x1024 resolution images. How did the author input the 512x512 resolution image? After I created LoRA_Sam, I tried to input a 512x512 resolution image, but it was prompted that the dimensions of the position embedding did not match.

hitachinsk commented 1 year ago

We use bilinear upsample to align the resolution.