===================================================================== Large-scale Fashion Recognition and Retrieval (DeepFashion) Dataset ===================================================================== ============================================= Category and Attribute Prediction Benchmark ============================================= -------------------------------------------------------- By Multimedia Lab, The Chinese University of Hong Kong -------------------------------------------------------- For more information about the dataset, visit the project website: http://mmlab.ie.cuhk.edu.hk/projects/DeepFashion.html If you use the dataset in a publication, please cite the papers below: @inproceedings{liu2016deepfashion, author = {Ziwei Liu, Ping Luo, Shi Qiu, Xiaogang Wang, and Xiaoou Tang}, title = {DeepFashion: Powering Robust Clothes Recognition and Retrieval with Rich Annotations}, booktitle = {Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}, month = June, year = {2016} } Please note that we do not own the copyrights to these images. Their use is RESTRICTED to non-commercial research and educational purposes. ======================== Change Log ======================== Version 1.0, released on 08/08/2016 Version 1.1, released on 22/12/2016, add landmarks annotations ======================== File Information ======================== - Images (Img/img.zip) 289,222 diverse clothes images. See IMAGE section below for more info. - Bounding Box Annotations (Anno/list_bbox.txt) bounding box labels. See BBOX LABELS section below for more info. - Fashion Landmark Annotations (Anno/list_landmarks.txt) fashion landmark labels. See LANDMARK LABELS section below for more info. - Category Annotations (Anno/list_category_cloth.txt & Anno/list_category_img.txt) clothing category labels. See CATEGORY LABELS section below for more info. - Attribute Annotations (Anno/list_attr_cloth.txt & Anno/list_attr_img.txt) clothing attribute labels. See ATTRIBUTE LABELS section below for more info. - Evaluation Partitions (Eval/list_eval_partition.txt) image names for training, validation and testing set respectively. See EVALUATION PARTITIONS section below for more info. ========================= IMAGE ========================= ------------ img.zip ------------ format: JPG --------------------------------------------------- Notes: 1. The long side of images are resized to 300; 2. The aspect ratios of original images are kept unchanged. --------------------------------------------------- ========================= BBOX LABELS ========================= ------------ list_bbox.txt ------------ First Row: number of images Second Row: entry names Rest of the Rows: <image name> <bbox location> --------------------------------------------------- Notes: 1. The order of bbox labels accords with the order of entry names; 2. In bbox location, "x_1" and "y_1" represent the upper left point coordinate of bounding box, "x_2" and "y_2" represent the lower right point coordinate of bounding box. Bounding box locations are listed in the order of [x_1, y_1, x_2, y_2]. --------------------------------------------------- ========================= LANDMARK LABELS ========================= ------------ list_landmarks_consumer2shop.txt ------------ First Row: number of images Second Row: entry names Rest of the Rows: <image name> <clothes type> <variation type> [<landmark visibility 1> <landmark location x_1> <landmark location y_1>, ... <landmark visibility 8> <landmark location x_8> <landmark location y_8>] --------------------------------------------------- Notes: 1. The order of landmark labels accords with the order of entry names; 2. In clothes type, "1" represents upper-body clothes, "2" represents lower-body clothes, "3" represents full-body clothes. Upper-body clothes possess six fahsion landmarks, lower-body clothes possess four fashion landmarks, full-body clothes possess eight fashion landmarks; 3. In variation type, "1" represents normal pose, "2" represents medium pose, "3" represents large pose, "4" represents medium zoom-in, "5" represents large zoom-in; 4. In landmark visibility state, "0" represents visible, "1" represents invisible/occluded, "2" represents truncated/cut-off; 5. For upper-body clothes, landmark annotations are listed in the order of ["left collar", "right collar", "left sleeve", "right sleeve", "left hem", "right hem"]; For lower-body clothes, landmark annotations are listed in the order of ["left waistline", "right waistline", "left hem", "right hem"]; For upper-body clothes, landmark annotations are listed in the order of ["left collar", "right collar", "left sleeve", "right sleeve", "left waistline", "right waistline", "left hem", "right hem"]. --------------------------------------------------- ========================= CATEGORY LABELS ========================= --------------- list_category_cloth.txt -------------- First Row: number of categories Second Row: entry names Rest of the Rows: <category name> <category type> --------------- list_category_img.txt -------------- First Row: number of images Second Row: entry names Rest of the Rows: <image name> <category label> --------------------------------------------------- Notes: 1. In category type, "1" represents upper-body clothes, "2" represents lower-body clothes, "3" represents full-body clothes; 2. The order of category labels accords with the order of category names; 3. In category labels, the number represents the category id in category names; 4. For the clothing categories, "Cape", "Nightdress", "Shirtdress" and "Sundress" have been merged into "Dress"; 5. Category prediction is treated as a 1-of-K classification problem. --------------------------------------------------- ========================= ATTRIBUTE LABELS ========================= --------------- list_attr_cloth.txt -------------- First Row: number of attributes Second Row: entry names Rest of the Rows: <attribute name> <attribute type> --------------- list_attr_img.txt -------------- First Row: number of images Second Row: entry names Rest of the Rows: <image name> <attribute labels> --------------------------------------------------- Notes: 1. In attribute type, "1" represents texture-related attributes, "2" represents fabric-related attributes, "3" represents shape-related attributes, "4" represents part-related attributes, "5" represents style-related attributes; 2. The order of attribute labels accords with the order of attribute names; 3. In attribute labels, "1" represents positive while "-1" represents negative, '0' represents unknown; 4. Attribute prediction is treated as a multi-label tagging problem. --------------------------------------------------- ========================= EVALUATION PARTITIONS ========================= ------------- list_eval_partition.txt ------------- First Row: number of images Second Row: entry names Rest of the Rows: <image name> <evaluation status> --------------------------------------------------- Notes: 1. In evaluation status, "train" represents training image, "val" represents validation image, "test" represents testing image; 2. Please refer to the paper "DeepFashion: Powering Robust Clothes Recognition and Retrieval with Rich Annotations" for more details. --------------------------------------------------- ========================= Contact ========================= Please contact Ziwei Liu (lz013@ie.cuhk.edu.hk) for questions about the dataset.