pmh9960 / GCDP

Official PyTorch implementation of "Learning to Generate Semantic Layouts for Higher Text-Image Correspondence in Text-to-Image Synthesis." (ICCV 2023)
https://pmh9960.github.io/research/GCDP/
41 stars 2 forks source link

Problem consultation #1

Open Gbone3176 opened 2 months ago

Gbone3176 commented 2 months ago

hello~ Our dataset does not have a corresponding semantic segmentation picture, so can we still use your network and weights for training? thanks!

pmh9960 commented 2 months ago

Thank you for your interest.

Unfortunately, our approach only supports image-label paired datasets. I’m not sure if this applies to your dataset, but it may not work well. Nevertheless, the advanced training techniques of joint diffusion models with unpaired datasets could be a promising research direction.

Best, Minho