The WebCaricature database is a large photograph-caricature dataset consisting of 6042 caricatures and 5974 photographs from 252 persons collected from the web. For each image in the dataset, 17 labeled facial landmarks are provided. As all the caricature images are collected from the web, the caricatures are of various artistic styles. Besides, the illumination conditions, pose, expressions, occlusions and age gap are not controlled for both the caricature and photo modalities.
We have provided four experimental settings, that are, restricted and unrestricted caricature verification, caricature to photo and photo to caricature identification. This offers a way for equal comparison of future works. Baseline performance under the four experimental settings is also provided.
Examples of Photographs and CaricaturesThe first column are photos and the next six columns are corresponding caricatures. The caricatures are of various artistic styles. There is a large intra-personal variations of caricatures.
- WebCaricature Dataset includes the following contents:
- Images of 6042 caricatures and 5974 photographs from 252 persons.
- 17 facial landmarks for each image
- 4 evaluation protocols
Restricted caricature face verification Unrestricted caricature face verification Caricature to photo (C2P) face identification Photo to caricature (P2C) face identificationn
Landmark Meaning Landmark Meaning 1 Contour (Top of hairline) 10 Right corner of left eye 2 Contour (Center of left ear) 11 Left corner of right eye 3 Contour (Chin)) 12 Right corner of right eye 4 Contour (Center of right ear) 13 Nose tip 5 Left corner of left eyebrow 14 Mouth upper lip top 6 Right corner of left eyebrow 15 Mouth left corner 7 Left corner of right eyebrow 16 Mouth lower lip bottom 8 Right corner of right eyebrow 17 Mouth right corner 9 Left corner of left eye
- Download instructions
- Please download the dataset release agreement, read it carefully and sign it. Notice, for students, the agreement should be signed by your supervisor.
- Please scan the signed document and send it to email@example.com and firstname.lastname@example.org. After the dataset application is approved, detailed download instructions will be sent back.
The database is made available only for non-commercial use. If you use this dataset, please cite the following paper:"WebCaricature: A Benchmark for Caricature Recognition"[bibtex] [arXiv] [webpage]
Jing Huo, Wenbin Li, Yinghuan Shi, Yang Gao, Hujun Yin,
British Machine Vision Conference, 2018.
"Variation Robust Cross-Modal Metric Learning for Caricature Recognition"[bibtex] [PDF] [webpage]
Jing Huo, Yang Gao, Yinghuan Shi, Hujun Yin,
ACM Multimedia Thematic Workshops, 2017: 340-348.
The database can also be used for caricature generation."CariGAN: Caricature Generation through Weakly Paired Adversarial Learning"[arXiv] [webpage]
Wenbin Li, Wei Xiong, Haofu Liao, Jing Huo, Yang Gao, Jiebo Luo,
arXiv preprint arXiv:1811.00445, 2018.
The dataset is released for research and educational purposes. We hold no liability for any undesirable consequences of using the dataset. All rights of the WebCaricature Dataset are reserved. Any person or organization is not permitted to distribute, redistribute, publish, copy or disseminate this dataset.