How to generate image-text pairs when i want to use other datasets such as ADE20K or medical segmentation dataset?

[s1mple-fake]

In your project, I didn't find any code for generating image-text pairs. But when you conduct the zero-transfer process on the Pascal context dataset, the dataset are seemly not have text label, so how to generate the text label?

[xvjiarui]

Hi @s1mple-fake

The texts are here. https://github.com/NVlabs/GroupViT/blob/main/segmentation/datasets/pascal_voc.py