📝 Diffusion Model Is Secretly a Training-Free Open Vocabulary Semantic Segmenter 🔭

"DiffSegmenter utilizes an off-the-shelf pre-trained text-conditioned diffusion model to generate cross-attention and self-attention maps, which are further refined and completed for final predictions." [gal30b+] 🤖 #CV

🔗 https://arxiv.org/abs/2309.02773v1 #arxiv

https://creative.ai/system/media_attachments/files/111/023/895/007/003/956/original/5b561a42ab2e2a2c.jpg

https://creative.ai/system/media_attachments/files/111/023/895/085/425/336/original/4c264b7d1607aeac.jpg

https://creative.ai/system/media_attachments/files/111/023/895/140/165/653/original/7600a030bd0b672e.jpg

https://creative.ai/system/media_attachments/files/111/023/895/194/206/978/original/996fb71f77b7dead.jpg