Before running any experiments, you need to ensure that the provided dataset is correctly placed within the data/
folder at the root of the project.
The directory structure of the data/
folder should look like this:
data/
│
├── bkai_polyp/
│ ├── anns/
│ │ ├── test.json
│ │ ├── train.json
│ │ └── val.json
│ ├── images/
│ └── masks/
│
├── [other dataset folders...]
│
└── kvasir_polyp/
├── anns/
│ ├── test.json
│ ├── train.json
│ └── val.json
├── images/
└── masks/
Each dataset folder (bkai_polyp
, busi
, camus
, etc.) contains three sub-directories: anns/
, images/
, and masks/
.
The anns directory contains prompt files (test.json
, train.json
, val.json
), while images/
and masks/
hold input images and target masks respectively.
The name of the mask files should be consistent with the mask_name
value given in the *.json
files; if not, you may need to rename them accordingly.
All of the masks must be binary segmentation images with 0
for background and 255
for foreground; if not, please process them accordingly.
Note: Since the datasets are of medical domain, you may need to comply with certain conditions before downloading and using them.