Description:

In order to provide benchmarking data for clothes material/category recognition in free con-figurations , we captured a high quality RGBD clothing dataset using a stereo head system. In our dataset,there are 50 items of clothing from 5 categories:  t-shits, shirts,  sweaters, jeans and towels, of which the material types are: cotton, jaconet, wool, denim, coarse cotton, respectively. Each item of clothing is captured in 21 different random configurations, resulting in a total of 1050 configurations. For each configuration the RGB image, depth map and the segmented mask are provided – both the RGB and depth data are 4 mega-pixels, 3264×4928. This is the first high-resolution free-configuration clothing dataset.

Objectives:

The recognition chanllenge is to find 3D visual representation that is robust to clothing's deformable configuration. As a consequence, algorithm should be able to recognise unknow configuration of unknown clothing.

Evaluation:

5-fold Cross-Validation on clothes level. This means the training examples and testing examples are from different items of clothing.

State-of-the-art Performance:

83.2% classification accuracy on Robot Head data, 64.2% on Kinect data.(Only 3D data is using)

Reference:

(1) Clothing Category Recognition in Free-Configurations with Application to Autonomous Clothes Sorting Robot

(2) Recognising the Clothing Categories from Free-Configuration using Gaussian-Process-Based Interactive Perception

Toolbox & Updates can be found in my personal page:

https://sites.google.com/site/clopemaclothesdataset/