SAM LoRA
Segment Anything fine-tuned for ring segmentation
This project tackles an issue about the segment anything model for segmenting rings and pairs of rings. To solve this problem I used a LoRA on the attention module. The dataset is handmade with a few training sets. I used the dice loss and the test set is based on this validation metric. I tested different dimensions for the LoRA matrices.
You can find all the details in the README of the repository: https://github.com/MathieuNlp/Sam_LoRA