--- license: other license_name: cc-by-nc-4.0-mixed-rights license_link: LICENSE task_categories: - image-classification language: - en tags: - image - datasets - computer-vision - image-retrieval - animal-re-identification - cat - fine-grained-recognition pretty_name: Individual Cats in the Wild size_categories: - 10K **MeowID: A Dual-Expert Retrieval System for Individual Cat Identification** > > Zhangchi Hu, Yi Shang, Haocheng Yang, Qiwei Hu, and Yuzheng Li (2026) - Project repository: https://github.com/RicePasteM/MeowID - Model weights: https://huggingface.co/RicePasteM/MeowID-Base ## Dataset structure ```text icw_split/ ├── train/ │ └── / │ └── .jpg ├── val/ │ └── / │ └── .jpg ├── test/ │ └── / │ └── .jpg ├── cats.csv ├── metadata.csv └── check_integrity.py ``` Each eight-digit `cat_id` denotes one identity. Identity directories are strictly disjoint across the training, validation, and test splits. | Split | Identities | Images | Images per identity | | --- | ---: | ---: | ---: | | Train | 18,877 | 77,094 | 3–19 (mean 4.08) | | Validation | 500 | 2,851 | 5–14 (mean 5.70) | | Test | 500 | 2,846 | 5–17 (mean 5.69) | | **Total** | **19,877** | **82,791** | **3–19 (mean 4.17)** | The complete directory occupies approximately 24 GB before repository-side compression. ## Construction The release was built through the following stages: 1. Public adoption and rescue profiles were collected from six platforms. 2. Automatic filtering retained cat images and profiles with sufficient views. 3. Cats were segmented, the exterior background was masked, and crops were tightened around the subject. 4. Near-duplicate observations and low-quality samples were removed through automated comparison and manual review. 5. Identities were assigned to mutually exclusive train, validation, and test splits. The identity distribution by source is: | Source | Identities | | --- | ---: | | PetFinder 2026 | 16,785 | | PetRescue | 2,881 | | 汪汪喵呜孤儿院 | 115 | | RSPCA Animals | 47 | | BC SPCA | 41 | | Verhuisdieren | 8 | Source names describe provenance, not endorsement of this dataset or its authors. ## Metadata Both CSV files are UTF-8 encoded with a byte-order mark. Use `encoding="utf-8-sig"` when reading them directly. ### `cats.csv` One row per identity: | Field | Description | | --- | --- | | `cat_folder` | Eight-digit identity directory name | | `source`, `source_name` | Source platform identifiers | | `animal_id` | Source-side animal/listing identifier | | `name`, `breed`, `gender` | Attributes reported by the source listing | | `referer_url` | Original public profile URL | | `metadata_json` | Preserved source and curation metadata | ### `metadata.csv` One row per image: | Field | Description | | --- | --- | | `cat_folder`, `image_filename` | Path key for the local JPEG | | `source`, `source_name`, `animal_id` | Source provenance | | `name`, `breed`, `gender` | Source-reported attributes | | `image_url`, `referer_url` | Original image and profile URLs | | `norm_x`, `norm_y`, `norm_w`, `norm_h` | Normalized crop bounding box | | `assignment_id` | Internal curation assignment identifier | | `metadata_json` | Preserved source and curation metadata | Source-reported fields may be incomplete, outdated, or inaccurate and should not be treated as verified biological labels. ## Loading the images The directory layout is compatible with common image-folder loaders. With Hugging Face Datasets: ```python from datasets import load_dataset dataset = load_dataset( "imagefolder", data_dir="/path/to/icw_split", drop_metadata=True, drop_labels=False, ) print(dataset) ``` The explicit `drop_metadata=True` is required because the root-level `metadata.csv` is ICW's provenance table rather than an ImageFolder metadata file with a `file_name` column. `drop_labels=False` keeps the identity inferred from each directory name. For identity retrieval, use directory names as identity labels and preserve the provided split boundaries. Do not merge identities across splits. ## Intended uses ICW is intended for non-commercial research and evaluation involving: - individual animal identification and re-identification; - fine-grained image retrieval; - route-aware face and whole-animal recognition; - representation learning under changes in pose, viewpoint, lighting, background, camera, and occlusion. The dataset is not intended for identifying people, inferring pet ownership, contacting source organizations or individuals, making automated animal-welfare decisions, or any commercial deployment. ## Limitations and responsible use - The dataset is drawn from adoption and rescue platforms rather than a random sample of the global cat population. - Geographic, platform, breed, age, health, photographic, and curation biases may affect model behavior. - Some identities have substantially more observations than others. - Listings and URLs can become outdated after collection. - Free-text source metadata can contain contact details or other incidental information. Do not use it to identify, profile, or contact people. - Image availability on a public webpage does not by itself imply that the image is freely relicensable. Users are responsible for evaluating legal, ethical, privacy, and source-site requirements in their jurisdiction and application. ## Data quality The included `check_integrity.py` verifies directory/CSV correspondence, identity disjointness, JPEG signatures, per-identity image counts, and bounding box ranges. The local release was validated on 17 August 2026: - 19,877 unique identity directories; - 82,791 corresponding metadata rows and JPEG files; - no identity overlap between train, validation, and test; - no missing, empty, or invalid-signature JPEG files; - all normalized bounding boxes within `[0, 1]` and with non-zero area. ## Licensing and source rights The dataset-specific selection, identity organization, split assignments, curation annotations, and original documentation contributed by the MeowID authors are licensed under the **Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0)**. See [`LICENSE`](LICENSE) and the [Creative Commons license page](https://creativecommons.org/licenses/by-nc/4.0/). This license applies only to material for which the MeowID authors hold the necessary rights. **Third-party photographs, source listing text, trademarks, and linked content are excluded from that license** and remain subject to their respective copyright holders' rights and source-platform terms. No endorsement by any source platform is implied. To request removal of an identity, image, source record, or stale contact information, open an issue in the [MeowID repository](https://github.com/RicePasteM/MeowID/issues) and include only the minimum path or source identifier needed to locate the record. ## Citation If you use ICW or MeowID in academic work, please cite: ```bibtex @article{hu2026meowid, title = {MeowID: A Dual-Expert Retrieval System for Individual Cat Identification}, author = {Hu, Zhangchi and Shang, Yi and Yang, Haocheng and Hu, Qiwei and Li, Yuzheng}, year = {2026} } ```