Download competition dataset from here
data_path/
├── train/
│ └── annotations/
│ | ├── train_ann.json
│ └──images/
| └── ...
|
├── test/
| └──annotations/
| ├── ... (仅包含问题(description)和答案(answer))
| └──images/
| └── ...
You can download the pre-trained weights of Mipha-3B, CLIP and Mask2Former from these links, and place them in the pretrained_model folder according to the following structure:
pretrained_model/
├── CLIP/siglip-so400m-patch14-384/
│ └── ...
├── mask2former/
│ └── model.pkl
├── mllm/Mipha-3B/
└── ...