🦖 Talk2DINO Demo

This demo allows you to perform open-vocabulary semantic segmentation on images using Talk2DINO.

How to use:

  1. Upload an image or select one from the example gallery.
  2. Enter a comma-separated list of class names you want to segment (e.g., pikachu, forest, road).
  3. Adjust optional parameters:
  • Model: choose between ViT-B and ViT-L
  • Apply PAMR: refine masks after initial prediction
  • Include Background: visualize background areas
  • Background Threshold: threshold for background intensity
  • Apply Background Cleaning: remove background noise when enabled
  1. Click Generate Segmentation to see the segmentation overlay.
Select Model
0 1