COCO-CMFD

A synthetic copy-move forgery dataset generated from MS-COCO 2017, with source/target-separated ground truth for copy-move forgery detection (CMFD).

Each sample takes one annotated COCO object, applies a mild affine transform, and pastes it elsewhere in the same image at a location that passes scene-plausibility checks (support surface, horizon band, perspective scale, occlusion). Ground truth is provided as a 3-class trimap, a binary mask, a 16 px patch-label grid, and a hard-negative mask of the scene's other objects.

Release v1.1 contains 8,245 samples in four parts, one per licence of the source photographs (see License). Release v1.0 (12,271 samples, CC BY 4.0) is withdrawn: 4,026 of its samples derived from photographs whose licence forbids adaptations.

Code, generation pipeline and quality tooling: https://github.com/harshitajain523/COCO-Copy-Move-Forgery-Dataset

Dataset details

Samples	8,245 (by 2,006 · by-nc 1,697 · by-sa 1,092 · by-nc-sa 3,450)
Source images	4,788 distinct COCO train2017 images (1–8 samples each, mean 1.7)
Categories	63 COCO categories
Resolution	native COCO, longer side 400–640 px
Target bounding box	2.5%–47.6% of image area (median 8.6%)
Transforms	scale 0.70–1.45, rotation ±10°, horizontal flip on 24.6%
Post-processing	none (no JPEG, noise or blur applied)

Fields

Field	Type	Description
image	Image	Forged image (PNG, lossless)
trimap	Image	0 background, 128 source, 255 target
binary_mask	Image	0/255, source ∪ target
hard_negatives	Image	0/255, all non-source annotated objects
coco_image_id, coco_license_id	int	Source photograph and its COCO licence id
license_name, license_url	string	Licence of the source photograph
flickr_url, coco_url	string	Where the source photograph is published, for attribution
modified	bool	Always true: every sample is an adaptation
patch_labels	list[int8]	Flattened 16 px grid: -1 ignore, 0 bg, 1 source, 2 target
patch_labels_shape	list[int]	[H/16, W/16] for reshaping patch_labels
gate_profile	string	strict, relaxed_v1, or relaxed_v2 (see below)
source_reused	int	1 if this source object is also the source of another sample from the same image
source_category_name	string	COCO category of the copied object
scale_factor, rotation_angle, flip	float/int	Transform applied to the copy
source_bbox, target_bbox	string	(x, y, w, h) tuples
placement_tier, horizon_tier	string	Which candidate pool / vertical constraint accepted the placement

24 generation metadata fields plus 7 licence fields; image_id refers to the originating COCO image.

Patch labels use a strict purity rule: a patch is labelled only if all 256 pixels share one class, otherwise -1. This keeps the copy-move boundary out of patch-level contrastive objectives.

Generation profiles

gate_profile	Samples	Source-object gates
strict	2,956	edge margin 15 px, isolation 5×5, bbox fill ≥ 0.25, paste ≥ 2% of image
relaxed_v1	115	edge 8 px, isolation 3×3, fill ≥ 0.20
relaxed_v2	5,174	as relaxed_v1, plus paste ≥ 1.2% of image

Scene-plausibility gates are identical across all three profiles; only source-object selectivity differs. Filter on gate_profile == "strict" for the most conservative subset.

Usage

from datasets import load_dataset
import numpy as np

ds = load_dataset("harshitajainn/coco-cmfd", "all", split="train")
# or one part: "by", "by-nc", "by-sa", "by-nc-sa"
s = ds[0]

image = s["image"]                              # PIL.Image
trimap = np.array(s["trimap"])
source_mask, target_mask = trimap == 128, trimap == 255
patches = np.array(s["patch_labels"], dtype=np.int8).reshape(
    s["patch_labels_shape"]
)

WebDataset tar shards are also published under wds/<part>/ for streaming-heavy pipelines; each sample key carries .png, .trimap.png, .binary.png, .hardneg.npy, .patches.npy and .json.

Verification

Measured on release v1.0 with the tooling in the GitHub repository; v1.1 is a subset and is re-checked by tools/verify_release.py:

• Mask alignment (60 samples): IoU between changed pixels and the labelled target region — mean 1.0000, min 0.9997, with 0 pixels modified outside any mask. Blending uses an inward-only 2 px feather, so the alpha ramp stays inside the labelled region and the ground truth is exact rather than approximate.
• Integrity: 61,355/61,355 referenced files present, 400 sampled pixel checks pass.
• Post-hoc plausibility audit: 12,190 clean / 81 rejected (0.66%).
• v1.1 release check, all 8,245 samples against the original COCO photographs: changed pixels vs. labelled target IoU mean 0.9999 (min 0.9300), 0 pixels changed outside any mask; every row's licence matches its part and the COCO annotation file.

Intended use and limitations

Intended for pretraining and evaluating copy-move forgery detectors, including source/target discrimination and patch-level contrastive objectives.

Limitations:

• Synthetic. Forgeries are algorithmically composited, not made by a human with intent to deceive. Detectors trained only on this data should be fine-tuned and evaluated on real-world benchmarks.
• No post-processing. Images are lossless PNG with no JPEG recompression, noise or blur. Robustness to those degradations is not exercised by this set.
• Mild transforms only (±10° rotation, 0.70–1.45× scale). Large rotations and extreme rescaling are out of distribution.
• Category imbalance follows COCO: person 7.8%, clock 7.7%, bird 7.2% are the most frequent of 63 categories.
• Source correlation: 2,723 samples reuse a source object that another sample from the same image also uses. Group by image_id when constructing splits to avoid leakage.
• Single copy-move per image; no multi-source or nested forgeries.

License and attribution

Each sample is an adaptation of one MS-COCO photograph, and COCO's photographs keep their Flickr owners' Creative Commons licences. The dataset is therefore released in four parts, each under the licence its source photographs allow:

Config	Source photographs	Licence
by	CC BY 2.0, no known copyright restrictions	CC BY 4.0
by-nc	CC BY-NC 2.0	CC BY-NC 4.0
by-sa	CC BY-SA 2.0	CC BY-SA 4.0
by-nc-sa	CC BY-NC-SA 2.0	CC BY-NC-SA 4.0

Photographs under NoDerivs licences are excluded. Every row carries its source photograph's licence, licence URL and Flickr URL, so attribution can be given per sample. Credit also the COCO Consortium (annotations, CC BY 4.0) and COCO-Stuff (labels used for placement, CC BY 4.0). Full terms: LICENSE-DATA and LICENSING.md in the GitHub repository.

Citation

@dataset{jain_coco_cmfd_2026,
  author    = {Jain, Harshita},
  title     = {{COCO-CMFD}: A Synthetic Copy-Move Forgery Dataset
               with Source/Target Ground Truth},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {TODO},
  url       = {https://github.com/harshitajain523/COCO-Copy-Move-Forgery-Dataset}
}

Please also cite MS-COCO:

@inproceedings{lin2014microsoft,
  title     = {Microsoft {COCO}: Common Objects in Context},
  author    = {Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and
               Hays, James and Perona, Pietro and Ramanan, Deva and
               Doll{\'a}r, Piotr and Zitnick, C Lawrence},
  booktitle = {ECCV},
  year      = {2014}
}