Kiminjo / MVTec-SAM-Validation

This repository aims to measure the zero-shot segmentation performance of Segment Anything Models (SAM) on Industrial defect data

MVTec-SAM-Validation

This repository aims to measure the zero-shot segmentation performance of Segment Anything Models (SAM) on Industrial defect data

Experiment configs

• prompt type
  • bounding box
  • point
• data
  • MVTec bottle
  • MVTec screw
  • MVTec wood data

RUN

1. Get data using dvc

cd data 
dvc pull mvtec.dvc 

2. Set MVTec's category in main.py

# main.py line 11
category = "wood" # Change here 

3. Run Code

python main.py

Then the results folder will be created and contain the inference result images.

4. (Option) Load previously learned results

dvc pull results.dvc