noureddinekhiati/diffseg-pytorch
1
DiffSeg — Unsupervised Zero-Shot Segmentation
Unofficial PyTorch implementation of DiffSeg from:
"Diffuse, Attend, and Segment: Unsupervised Zero-Shot Segmentation using Stable Diffusion" Tian et al., CVPR 2024
Upload any image — no text prompt, no labels, no training needed. The model uses self-attention maps from a frozen Stable Diffusion 1.5 UNet to segment objects.
Author: KHIATI Rezkellah Noureddine — GitHub