Prepare Datasets

May 6, 2026 · View on GitHub

We provide instructions for preparing the dataset for LiteLVLM.

Download Datasets

For Referring Expression Segmentation, we adopt the COCO Referring Expression Comprehension datasets: RefCOCO, RefCOCO+, and RefCOCOg.

Download links and directory structure:

images: COCO2014 - train2014
RefCOCO
RefCOCO+
RefCOCOg

Download the data and orgnaize as follows:

data
├── Refer_Segm
│   ├── refcoco
│   │   ├── instances.json 
│   │   ├── refs(google).p
│   │   ├── refs(unc).p
│   ├── refcoco+
│   ├── refcocog
│   ├── coco_2014
│   │   ├── train2014
│   │   │   ├── COCO_train2014_000000000009.jpg
│   │   │   ├── COCO_train2014_000000000025.jpg