Skip to content

Latest commit

 

History

History
157 lines (120 loc) · 3.95 KB

DATASET.md

File metadata and controls

157 lines (120 loc) · 3.95 KB

Please prepare dataset in the following format.

For easier use, we suggest to use the datasets we have pre-processed in Hugging Face OMG-Seg dataset and OMG-LLaVA.

OMG-Seg datasets

[PS/IS] COCO dataset

The default setting as mmdetection.

├── coco
│   ├── annotations
│   │   ├── panoptic_{train,val}2017.json
│   │   ├── instance_{train,val}2017.json
│   ├── train2017
│   ├── val2017
│   ├── panoptic_{train,val}2017/  # png annotations

[PS] Cityscapes dataset

Please use the scripts in ext/cityscapes_scripts/createPanopticImgs.py to generate COCO-style cityscape panoptic segmentation format.

python ext/cityscapes_scripts/createPanopticImgs.py --dataset-folder ./data/cityscapes --output-folder ./data/cityscapes
├── cityscapes
│   ├── annotations
│   │   ├── cityscapes_panoptic_train_trainId.json  # panoptic json file 
│   │   ├── cityscapes_panoptic_val_trainId.json 
│   │   ├── cityscapes_panoptic_train_trainId # panoptic png file
│   │   ├── cityscapes_panoptic_val_trainId # panoptic png file
│   ├── leftImg8bit # training images
│   ├── gtFine # origin gt files 
│   │   ├──
│   │   

[VIS] Youtube-VIS (2019/2021) dataset

Use the scripts tools/dataset_convert/vis_to_coco.py to convert origin json to COCO-style.

python tools/dataset_convert/vis_to_coco.py -i ./data/youtubevis2019 --version 2019
python tools/dataset_convert/vis_to_coco.py -i ./data/youtubevis2021 --version 2021

The final results are shown here:

├── youtubevis2019
│   ├── annotations
│   │   ├── youtube_vis_2019_train.json
│   │   ├── youtube_vis_2019_valid.json
│   ├── train
│   │   ├──JPEGImages
│   │   │   ├──video folders
│   ├── valid
│   │   ├──JPEGImages
│   │   │   ├──video folders
├── youtubevis2021
│   ├── annotations
│   │   ├── youtube_vis_2021_train.json
│   │   ├── youtube_vis_2021_valid.json
│   ├── train
│   │   ├──JPEGImages
│   │   │   ├──video folders
│   ├── valid
│   │   ├──JPEGImages
│   │   │   ├──video folders

[VPS] VIPSeg dataset

Download the origin dataset from the official repo.
Following official repo, we use resized videos for training and evaluation (The short size of the input is set to 720).

├── VIPSeg
│   ├──  imgs
│   │   ├── 1241_qYvEuwrSiXc
        │      ├──*.jpg
│   ├──  panomasks 
│   │   ├── 1241_qYvEuwrSiXc
        │      ├──*.png
│   ├──  panomasksRGB 

[SS/PS] ADE dataset

The default setting as mmdet, note that please use our pre-processed ADE annotations.

├── ade
│   ├──  ADEChallengeData2016
│   │   ├── images/
│   │   ├── annotations/
│   │   ├── ade20k_panoptic_train/
│   │   ├── ade20k_panoptic_val/
│   │   ├── ade20k_panoptic_train.json
│   │   ├── ade20k_panoptic_val.json

[VOS] DAVIS dataset

Please download DAVIS datasets as default.

Finally, link the download the dataset into the data folder as

root
├── ext
├── figs
├── seg
├── omg_llava
├── tools
├── data
│   ├──coco
│   ├──ade
│   ├──cityscapes
│   ├──VIPSeg
│   ├──youtube_vis_2019
│   ├──youtube_vis_2021
│   ├──DAVIS

OMG-LLaVA datasets

Please download OMG-LLaVA dataset from HuggingFace webpage.

├── data
│   ├──glamm_data
│   ├──llava_data
│   ├──mdpv_point
│   ├──ref_seg
│   ├──region_caption
│   ├──semantic_seg