iMaterialist Competition at FGVC6 (Workshop CVPR 2019)

Date:

This competition is part of the Fine-Grained Visual Categorization FGVC6 workshop at the Computer Vision and Pattern Recognition Conference CVPR 2020. [Official competition]

Acknowledgments: Google AI, CVDF, Samasource and Fashionpedia.

You can find my code at my github

Overview

Visual analysis of clothing is a topic that has received increasing attention in recent years. Being able to recognize apparel products and associated attributes from pictures could enhance the shopping experience for consumers, and increase work efficiency for fashion professionals.

The organizers present a new clothing dataset with the goal of introducing a novel fine-grained segmentation task by joining forces between the fashion and computer vision communities. The proposed task unifies both categorization and segmentation of rich and complete apparel attributes, an important step toward real-world applications.

Solution

Mask R-CNN, which is based on top of Faster R-CNN. Also tried U-Net for segmentation.

The implementation was based on this Pytorch tutorial. And obviously, using the magic words import torchvision

Details

  • Mask R-CNN pretrained on COCO (https://www.github.com/matterport/Mask_RCNN)
  • Training using 1 Tesla P100 (8-9hr)
  • Resnet101 Backbone
  • Images resized to 512x512 and using Horizontal Flip Augmentations

Demostration


example


Public Demo video by DataScience (Youtube channel)