search

NVlabs / ODISE

Official PyTorch implementation of ODISE: Open-Vocabulary Panoptic Segmentation with Text-to-Image Diffusion Models [CVPR 2023 Highlight]
https://arxiv.org/abs/2303.04803
Other
845 stars 45 forks source link

Should the input image be in RGB or BGR? #41

Open FunnyDragonK opened 8 months ago

FunnyDragonK commented 8 months ago

Thanks for the excellent work in open source. The run_on_image function says the input image should be in BGR order. But in the demo code, the input image is in RGB mode. So, I'm unsure about which mode would yield better results.

Additionally, I found that the ODISE(Lable) model doesn't recognize the "poles". What could be the reason? Is the prompt "poles" incorrect?