Tag-My-Outfit
Tag-My-Outfit: a pipeline to classify fashion objects
Tag-My-Outfit classifies clothing parts within a set of categories, as well as attributes for the clothes. It contains a multiclass and a multilabel classifier.
The pipeline reads a folder of a set of files, classifies them, and provides a visualization of the images as well as the description of the predicted category and attributes.
Hardware architecture: X64
Instructions to run: Download the zip file and unzip it. Then, run the following command in the unzipped folder that contains the file
docker-compose.yml: >> docker-compose up
To use this docker image, you can build a client with the example (in python) provided in: https://github.com/sipg-isr/tag_my_outfit_client
Additional information: The Tag My Outfit service predicts the category and the attributes of a clothing item present in the given image. The prediction model is the Visual Semantic Attention Model (VSAM) from [1], a compact framework with guided attention for multi-label classification in the fashion domain. This model is supervised by automatic pose extraction creating a discriminative feature space. Specifically, VSAM regularizes the VGG-16 backbone network with ground-truth heatmaps with relevant joints. VSAM performs on pair with previous works on the DeepFashion dataset, even without using any landmark annotations. The code made available through this service allows performing inference with the VSAM model pretrained on the DeepFashion dataset [2].
For more details check: http://physicalai.isr.tecnico.ulisboa.pt/tagmyoutfit.html
This AI asset was developed by the AI4EU Physical AI team from ISTR
References:
[1] Beatriz Quintino Ferreira, Joao P. Costeira, Ricardo G. Sousa, Liang-Yan Gui, Joao P. Gomes, “Pose Guided Attention for Multi-label Fashion Image Classification”, ICCV Workshop Computer Vision for Fashion, Art and Design, ICCV 2019
[2] DeepFashion dataset website http://mmlab.ie.cuhk.edu.hk/projects/DeepFashion.html