---
title: CORNE
canonical_url: "https://www.modelscope.cn/datasets/ZhouqmR/CORNE"
md_url: "https://www.modelscope.cn/datasets/ZhouqmR/CORNE.md"
repository: ZhouqmR/CORNE
last_updated: 2026-07-28
license: apache-2.0
storage_size: "711 GB"
downloads: 31736
stars: 1
---

# CORNE

> CORNE - ZhouqmR 在 ModelScope 开源的数据集。CORNE is an effect-aware paired dataset for object removal. It contains 287,012 source/background pairs constructed from the single-edit addition and removal subset of NHR-Edit using the Semantic-Aware Visual Processing…

ZhouqmR/CORNE 是 ModelScope 魔搭社区上的数据集，存储大小 711 GB，采用 apache-2.0 许可。

- **Repository**: ZhouqmR/CORNE
- **License**: apache-2.0
- **Storage size**: 711 GB
- **Downloads**: 31736
- **Stars**: 1
- **Last updated**: 2026-07-28

Source: https://www.modelscope.cn/datasets/ZhouqmR/CORNE

---

# CORNE

[Paper](https://arxiv.org/abs/2606.28094) | [Code](https://github.com/Zhouqm-Git/osor)

CORNE is an effect-aware paired dataset for object removal. It contains **287,012** source/background pairs constructed from the single-edit addition and removal subset of [NHR-Edit](https://huggingface.co/datasets/iitolstykh/NHR-Edit) using the Semantic-Aware Visual Processing (SAVP) pipeline introduced in OSOR.

## Statistics

| Item | Number |
| --- | ---: |
| NHR-Edit samples considered | 680,088 |
| Single-edit addition/removal samples | 427,929 |
| Retained CORNE pairs | 287,012 |
| Effect-heavy pairs used in OSOR Phase II | 67,726 |

## Data Contents

The released files are organized as:

```text
CORNE/
├── shot/          # Source images containing the target object
├── bg/            # Paired object-free backgrounds
├── mask-check/    # Effect-aware removal masks
├── mask_sam/      # Object-core masks
└── *.txt          # Sample lists and split metadata
```

Images with the same relative filename form a training pair. The metadata files define the subsets used by the released training code.

## Construction

1. Select single-edit addition/removal triplets from NHR-Edit and order each pair as source image and object-free background.
2. Estimate changed regions from the aligned image pair and verify their semantic consistency with the edit instruction.
3. Obtain object-core masks with SAM2.
4. Fuse object and residual-effect regions to produce the effect-aware removal mask.

This filtering and mask construction process is described in the OSOR paper and supplementary material.

## Provenance and License

CORNE is derived from NHR-Edit, which is released under the [Apache License 2.0](https://huggingface.co/datasets/iitolstykh/NHR-Edit/blob/main/LICENSE). CORNE, including the OSOR-produced masks and metadata distributed with it, is released under the Apache License 2.0. Users should retain the original NHR-Edit attribution and comply with its terms.

## Citation

If you use CORNE, please cite both OSOR and NHR-Edit:

```bibtex
@inproceedings{zhou2026osor,
  title     = {OSOR: One-Step Diffusion Inpainting for Effect-Aware Object Removal},
  author    = {Zhou, Qinming and Sun, Chenxi and Kong, Deyang and He, Junhao and Tang, Xiangheng and Yu, Peike and Wu, Haotian and Cao, Leilei and Zhang, Linfeng},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026},
  url       = {https://arxiv.org/abs/2606.28094}
}

@article{Layer2025NoHumansRequired,
  arxivId      = {2507.14119},
  author       = {Maksim Kuprashevich and Grigorii Alekseenko and Irina Tolstykh and Georgii Fedorov and Bulat Suleimanov and Vladimir Dokholyan and Aleksandr Gordeev},
  title        = {NoHumansRequired: Autonomous High-Quality Image Editing Triplet Mining},
  year         = {2025},
  eprint       = {2507.14119},
  archivePrefix = {arXiv},
  primaryClass = {cs.CV},
  url          = {https://arxiv.org/abs/2507.14119},
  journal      = {arXiv preprint arXiv:2507.14119}
}
```