Datasets:

InSAI-Lab
/

SynPASS

license: apache-2.0
task_categories:
  - image-segmentation
task_ids:
  - semantic-segmentation
pretty_name: SynPASS
tags:
  - panoramic-vision
  - autonomous-driving
  - semantic-segmentation
  - synthetic
  - 360-degree
  - cvpr-2022

SynPASS: Synthetic Panoramic Semantic Segmentation Benchmark

Official Dataset Repository for the SynPASS benchmark (Trans4PASS / Trans4PASS+).

GitHub Repository: https://github.com/InSAI-Lab/Trans4PASS
Paper (Trans4PASS, CVPR 2022): arXiv:2203.01452
Paper (Trans4PASS+, arXiv 2022): arXiv:2207.11860
Model Checkpoints: InSAI-Lab/Trans4PASS

Dataset Description

The SynPASS dataset contains 9,080 panoramic images (1024x2048) across 22 semantic categories. The dataset covers diverse scenarios:

Weather: Cloudy, Foggy, Rainy, Sunny
Lighting: Day-time, Night-time

Statistics

Condition	Cloudy	Foggy	Rainy	Sunny	Total
Split (train/val/test)	1420/420/430	1420/430/420	1420/430/420	1440/410/420	5700/1690/1690
Split (day/night)	1980/290	1710/560	2040/230	1970/300	7700/1380
Total Frames	2270	2270	2270	2270	9080

File Structure

The dataset archive SynPASS.tar.gz (~4.8 GB) unpacks into:

SynPASS/
├── img/
│   ├── cloud/
│   ├── fog/
│   ├── rain/
│   └── sun/
└── semantic/
    ├── cloud/
    ├── fog/
    ├── rain/
    └── sun/

Quick Start & Download

Using the Hugging Face CLI:

# Download dataset archive
hf download InSAI-Lab/SynPASS SynPASS.tar.gz --type dataset --local-dir ./datasets

# Extract archive
cd ./datasets
tar -zxvf SynPASS.tar.gz

Using Python (huggingface_hub):

from huggingface_hub import hf_hub_download

dataset_path = hf_hub_download(
    repo_id="InSAI-Lab/SynPASS",
    repo_type="dataset",
    filename="SynPASS.tar.gz",
    local_dir="./datasets"
)

Citation

@inproceedings{zhang2022bending,
  title={Bending Reality: Distortion-aware Transformers for Adapting to Panoramic Semantic Segmentation},
  author={zhang, Jiaming and Yang, Kailun and Ma, Chaoxiang and Rei{\ss}, Simon and Peng, Kunyu and Stiefelhagen, Rainer},
  booktitle={CVPR},
  pages={16917--16927},
  year={2022}
}

@article{zhang2022behind,
  title={Behind Every Domain There is a Shift: Adapting Distortion-aware Vision Transformers for Panoramic Semantic Segmentation},
  author={Zhang, Jiaming and Yang, Kailun and Shi, Hao and Rei{\ss}, Simon and Peng, Kunyu and Ma, Chaoxiang and Fu, Haodong and Wang, Kaiwei and Stiefelhagen, Rainer},
  journal={arXiv preprint arXiv:2207.11860},
  year={2022}
}

License

This dataset is released under the Apache-2.0 License.