metadata

license: cc-by-4.0

task_categories: - object-detection license: mit tags: - computer vision - amodal-tracking - object-tracking - amodal-perception

Segment-Object Dataset

This dataset is collected from LVIS and COCO. We employed the segments in this dataset to implement PasteNOcclude augmentation proposed in Tracking Any Object Amodally.

📙 Project Page | 💻 Code | 📎 Paper Link | ✏️ Citations

Contact: 🙋🏻‍♂️Cheng-Yen (Wesley) Hsieh

Dataset Download

git lfs install
git clone git@hf.co:datasets/chengyenhsieh/TAO-Amodal-Segment-Object-Large

After downloading this dataset, check here to see how to train our Amodal Expander with PasteNOcclude.

📚 Dataset Structure

The dataset should be structured like this:

   TAO-Amodal-Segment-Object-Large
    ├── train-2017
    │    ├── OOOOOO_XXX.jpg
    └── segment_object.json

📚 File Descriptions

File Name	Description
segment_object.json	Mask annotations of each segment object

Citation

@misc{hsieh2023tracking,
    title={Tracking Any Object Amodally},
    author={Cheng-Yen Hsieh and Tarasha Khurana and Achal Dave and Deva Ramanan},
    year={2023},
    eprint={2312.12433},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}