We recommend you to open this page on GitHub, so you can see the corresponding scripts and configs.

Pairwise postprocessing (re-ranking)

If you are new to Pipelines idea, you can start with general overview.

This Pipeline is based on the following study, completed by the OML’s team:

STIR: Siamese Transformer for Image Retrieval Postprocessing 

In this work, we first construct a baseline model trained with triplet loss with hard negatives mining that performs at the state of the art level but remains simple. Second, we introduce a novel approach for image retrieval postprocessing called Siamese Transformer for Image Retrieval (STIR) that re-ranks several top outputs in a single forward pass. Unlike previously proposed Reranking Transformers, STIR does not rely on global/local feature extraction and directly compares a query image and a retrieved candidate on pixel level with the usage of attention mechanism. The resulting approach defines a new state of the art on standard image retrieval datasets: Stanford Online Products and DeepFashion In-shop.

OPEN INTERACTIVE DEMO | demo’s repository

I. Train & validate a feature extractor

DeepFashion Inshop

python train_extractor.py dataset_root=data/InShop/ logs_root=logs/InShop
python validate_extractor.py dataset_root=data/Inshop/ weights=extractor_inshop.ckpt

Stanford Online Products

python train_extractor.py dataset_root=data/SOP/ logs_root=logs/SOP
python validate_extractor.py dataset_root=data/SOP/ weights=extractor_sop.ckpt

II. Train & validate a postprocessor