https://github.com/neka-nat/pytorch-hdml

Hardness-Aware Deep Metric Learning (CVPR2019) in pytorch
https://github.com/neka-nat/pytorch-hdml

cars196 cub200 cvpr2019 deep-learning deep-metric-learning pytorch triplet-loss

Last synced: 7 months ago
JSON representation

Hardness-Aware Deep Metric Learning (CVPR2019) in pytorch

• Host: GitHub
• URL: https://github.com/neka-nat/pytorch-hdml
• Owner: neka-nat
• License: mit
• Created: 2019-08-20T13:13:48.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2021-05-02T14:06:10.000Z (over 4 years ago)
• Last Synced: 2025-03-18T22:35:51.620Z (7 months ago)
• Topics: cars196, cub200, cvpr2019, deep-learning, deep-metric-learning, pytorch, triplet-loss
• Language: Python
• Homepage:
• Size: 20.6 MB
• Stars: 28
• Watchers: 2
• Forks: 7
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Hardness-Aware Deep Metric Learning

This is an unofficial implementation of ["Hardness-Aware Deep Metric Learning" (CVPR 2019 Oral)](https://arxiv.org/abs/1903.05503) in Pytorch.

## Installation

```

cd pytorch-hdml

pip install pipenv

pipenv install

```

## Download dataset

```

cd data

python cars196_downloader.py

python cars196_converter.py

```

## Train CARS196 dataset

Execute a training script. 

When executed, the tensorboard log is saved.

```

pipenv shell

python train_triplet.py

```

## Result triplet HDML

### CARS196 result on training(99 classes, 30000 iterations)

#### Loss

![loss](assets/triplet_loss.png)

#### t-SNE

![tsne](assets/triplet_train_tsne.gif)

### CARS196 result on testing(97 classes)

#### t-SNE

![tsne](assets/triplet_test_tsne.gif)

## Todo

- [ ] Implementation of Npair loss HDML

## Reference

Official tensorflow implementation https://github.com/wzzheng/HDML