Skip to content

Latest commit

 

History

History
13 lines (9 loc) · 857 Bytes

README.md

File metadata and controls

13 lines (9 loc) · 857 Bytes
Raw

Input/Inferenced images are 640x640 and upsample=True, otherwise must scale the masks. I havent played with non 640x640 images or tested it but will get around to it eventually

How It Works:

  1. Add --save-masks to segment/predict.py argument
  2. During inference, only the masks without the bounded boxes are saved to masks/ in the format of in a .txt file named as the input image.
  • Polygon coordinates are normalized to image size and saved as their contour coordinates

To review the masks

in review_predictions/ I made a GUI to visualize and buffer the images in the root directory with their bounded boxes and masks from the original predict.py output I was too lazy to use the masks from masks/ but will add these + other features eventually For info on the review_predictions usages I added a readme in there