CamoPrompt: Textual Prompts for Camouflaged Object Detection (COD)

CamoPrompt

Metadata

Entries: 4,040

Benchmarks covered: COD10K-Train and CAMO-Train

License (text annotations): CC BY 4.0

Images: No images are redistributed; please obtain originals from COD10K/CAMO under their licenses.

OverviewCamoPrompt contains 4,040 textual prompts that enrich the training splits of two public camouflaged-object-detection benchmarks (COD10K-Train and CAMO-Train). Each prompt is an object-centric, single-sentence English description focusing on target attributes (shape, fins/spines, body plan, markings, color/texture).

JSON schema

image_name (str): File name used in the original dataset

category (str): Coarse category label (e.g., frog, seahorse)

nl_description (str): One-sentence, object-focused description (color, texture, posture, etc.)

Usage

Pair image_name with the corresponding file from the original COD10K/CAMO datasets(No images are included in this release.

Use category as a coarse label; use nl_description for text-guided training/evaluation.

CitationPlease cite this record’s DOI (shown on the landing page) when you use the prompts, and also cite the original datasets:

Le, T.-N.; Nguyen, T. V.; Nie, Z.; Tran, M.-T.; Sugimoto, A. Anabranch Network for Camouflaged Object Segmentation. Computer Vision and Image Understanding, 184 (2019): 45–56. https://doi.org/10.1016/j.cviu.2019.04.006

Fan, D.-P.; Ji, G.-P.; Sun, G.-L.; Cheng, M.-M.; Shen, J.-B.; Shao, L. Camouflaged Object Detection. CVPR (2020): 2774–2784. https://doi.org/10.1109/CVPR42600.2020.00285