Computer Vision for Fashion at ICCV 2017

It took a while, but fashion has finally landed in the computer vision community. On one side, there are more and more industry specific workshops like the iMaterialist challenge by Google at CVPR or the Fashion workshop by Zalando at ICCV this year. Fashion related publications are also gaining their importance as main topics at these conferences, . Following the proven scientific method of Pont-Tuset et al., we visualized the percentage of fashion related papers at ICCV/CVPR in the past few years. To our great delight, fashion related papers steadily have gained traction!


As a researcher at Fashwell, it is intriguing for me to see how the key problem of product recognition unifies vast variety of different aspects of machine learning and computer vision. This starts with image tagging with noisy, hierarchical labels, object localization, representation/metric learning and finally ends with speeding up data annotation.

ICCV Awards

In the main conference track, object localization is still one of the hottest topics. This was highlighted by ‘The Best Paper’ and ‘The Best Student Paper’ awards for Mask R-CNN and Focal Loss, from the detection all-star team Girshick, He and Dollár (et al.). Other trending topics at the conference were GANs, in all their beauty. In particular, we were drawn to the work that generated outfits based on text or human body pose. Metric and representation learning were also a prevalent topic, either just with images or with joint text and image representations. Especially Kristen Grauman’s line of work, which presented an elegant way to mine styles and predict trends, caught our attention.

We completed our yearly ICCV visit by attending the fashion workshop on the last day. This is where various researchers from universities and the industry presented their recent work. Also here, object localization, image generation and representation learning remain some of the strongest foci of active applied research.

And by the way, if you are a computer vision researcher interested in solving product recognition: We are hiring!