This repository contains model and weights from paper "Fine-tuning CNN Image Retrieval with No Human Annotation", ResNet101 fine-tuned network with GeM pooling and contrastive loss, ported from MatConvNet to Pytorch. Transfer was done with the help of pytorch-mcn - tool for importing trained MatConvNet models into Pytorch.

In order to run the script, please, download weights from the Google Drive and store them in the weights folder.