may i test my images? not train ,how to do ?

[ZiqinZhou66]

I appreciate your interest in our work.

Yes, you can test your own images by the following steps:

1. Download our pre-trained model
2. Preparing your dataset in MMSeg format
3. Preparing a data loader file and put it in 'configs/base/datasets/dataloader/'
4. Preparing a data config file and put it in 'configs/base/datasets/'
5. Preparing target testing config file and put it in 'configs/yourData/'
6. run test.py with your config file

[aliman80]

Hi many thanks for your quick response, I did the following before the experiment;

1. In configs/base/datasets/cocostuff_512x512.py on line no 3 i updated the path data_root = 'Path/to/data/coco_stuff164k'
2. then in configs/coco/vpt_seg_fully_vit-b_512x512_80k_12_100_multi.py i updated the path pretrained = 'Path/to/pretrained/ViT-B-16.pt' with your pretrained model.
3. Then in https://github.com/ZiqinZhou66/ZegCLIP/blob/main/configs/_base_/models/zegclip.py line 9 pretrained='Path/to/pre trained/RN50.pt' I downloaded weights from the internet and updated this path
4. then on https://github.com/ZiqinZhou66/ZegCLIP/blob/main/configs/_base_/datasets/dataloader/coco_stuff.py line 108 i changed the suffix to png in place of _labelTrainIds.png.
5. Then i ran the validation experiment.