This repository contains the Pinterest shop the look data used in the following paper:

Wang-Cheng Kang, Eric Kim, Jure Leskovec, Charles Rosenberg, Julian McAuley (2019). Complete the Look: Scene-based Complementary Product Recommendation. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR'19)

Please cite our paper if you use the datasets.

We provide the two datasets for fashion and home respectively. Each dataset contains the scene-product pairs in the following format:

Example (fashion.json):
{
    "product": "0027e30879ce3d87f82f699f148bff7e", 
    "scene": "cdab9160072dd1800038227960ff6467", 
    "bbox": [
        0.434097, 
        0.859363, 
        0.560254, 
        1.0
    ]
}

The scene and product images are represented by a signature, and the corresponding image URL can be obtained via:

def convert_to_url(signature):
    prefix = 'http://i.pinimg.com/400x/%s/%s/%s/%s.jpg'
    return prefix % (signature[0:2], signature[2:4], signature[4:6], signature)

Also, product category is also provided (e.g. fashion-cat.json). Please refer to the paper for more statistics and details. You could contact me (wckang#ucsd.edu) if there is any question.

The bounding box is represented by four numbers: (left, top, right, bottom). All the numbers are normalized. The coordinates of the top left corner is (0,0).

• Add an illustrative example.