https://github.com/kunhe/FastAP-metric-learning

Code for CVPR 2019 paper "Deep Metric Learning to Rank"
https://github.com/kunhe/FastAP-metric-learning

cvpr deep-metric-learning learning-to-rank matconvnet metric-learning pytorch

Last synced: 5 months ago
JSON representation

Code for CVPR 2019 paper "Deep Metric Learning to Rank"

• Host: GitHub
• URL: https://github.com/kunhe/FastAP-metric-learning
• Owner: kunhe
• License: other
• Created: 2019-06-06T07:36:00.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2021-09-21T01:48:46.000Z (about 4 years ago)
• Last Synced: 2024-08-04T03:08:30.229Z (about 1 year ago)
• Topics: cvpr, deep-metric-learning, learning-to-rank, matconvnet, metric-learning, pytorch
• Language: MATLAB
• Homepage:
• Size: 102 KB
• Stars: 95
• Watchers: 6
• Forks: 16
• Open Issues: 2
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# FastAP: Deep Metric Learning to Rank

This repository contains implementation of the following paper:

[Deep Metric Learning to Rank](http://openaccess.thecvf.com/content_CVPR_2019/html/Cakir_Deep_Metric_Learning_to_Rank_CVPR_2019_paper.html)


[Fatih Cakir](http://cs-people.bu.edu/fcakir/)\*, [Kun He](http://cs-people.bu.edu/hekun/)\*, [Xide Xia](https://xidexia.github.io), [Brian Kulis](http://people.bu.edu/bkulis/), and [Stan Sclaroff](http://www.cs.bu.edu/~sclaroff/) (*equal contribution)


IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019

![](fastap_github.jpg)

## Other Implementations

[FastAPLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#fastaploss) from [pytorch-metric-learning](https://github.com/KevinMusgrave/pytorch-metric-learning)

## Usage 

* **Matlab**: see `matlab/README.md`

* **PyTorch**: see `pytorch/README.md`

## Datasets

* Stanford Online Products

  * Can be downloaded [here](http://cvgl.stanford.edu/projects/lifted_struct/)

* In-Shop Clothes Retrieval

  * Can be downloaded [here](http://mmlab.ie.cuhk.edu.hk/projects/DeepFashion.html)

* PKU VehicleID

  * Please request the dataset from the authors [here](https://pkuml.org/resources/pku-vehicleid.html)

## Reproducibility

* We provide trained [MatConvNet](https://www.vlfeat.org/matconvnet/quick/) models and experimental logs for the results in the paper. These models were used to achieve the results in the tables.

* The logs also include parameters settings that enable one to re-train a model if desired. It also includes evaluation results with model checkpoints at certain epochs.

    * Table 1: Stanford Online Products

        * FastAP, ResNet-18, M=256, Dim=512: [[model @ epoch 20](https://drive.google.com/file/d/1sPCG34rV4Bqf0aWF7GrFIDUK7DGjcaB5/view?usp=sharing), [log](https://drive.google.com/open?id=14m3fHgeZu8MIAePFHXe141R60KRwH1d8)]

        * FastAP, ResNet-50, M=96, Dim=128: [[model @ epoch 30](https://drive.google.com/open?id=1yGUVTskdERdLeF85GP-lLRnwS0KhpvvL), [log](https://drive.google.com/open?id=1A0G1aUBS7URotInbT7eBbys4xfARvCCe)]

        * FastAP, ResNet-50, M=96, Dim=512: [[model @ epoch 28](https://drive.google.com/file/d/14yEyAYhGzNygBBn8r2RcZut_Ye14mJoK/view?usp=sharing), [log](https://drive.google.com/open?id=19mpLn1OqA2nqpMZtvZ3GvFk_VppOOPTc)]

        * FastAP, ResNet-50, M=256, Dim=512: [[model @ epoch 12](https://drive.google.com/open?id=1WfV1ArXHG4oksHGE8DZRDwsxupoO60sD), [log](https://drive.google.com/open?id=1shvC5qB8O0l6vH1qa2SG1oxX8_jM1gdi)]

    * Table 2: In-Shop Clothes

        * FastAP, ResNet-18, M=256, Dim=512: [[model @ epoch 50](https://drive.google.com/open?id=1ZZ-Fpx9uPkRL-QXL-8-RcROjQVOLcbr5), [log](https://drive.google.com/file/d/1osxoHsMy11v-kvUNTuRG3luhsxMMn78B/view?usp=sharing)]

        * FastAP, ResNet-50, M=96, Dim=512: [[model @ epoch 40](https://drive.google.com/open?id=1PyiHog7fJp_InvqdAO0dzyJDRMNvAXxm), [log](https://drive.google.com/open?id=14IPgDfkbKo9PnrgMFFRDSIBRW1xwRYs5)]

        * FastAP, ResNet-50, M=256, Dim=512: [[model @ epoch 35](https://drive.google.com/open?id=1T5IynM63YqnGslnMGppJsmtJdHIWamJv), [log](https://drive.google.com/open?id=1oud9i87FTJE7Ei636bjxqgBXpahysRKK)]

    * Table 3: PKU VehicleID

        * FastAP, ResNet-18, M=256, Dim=512: [[model @ epoch 50](https://drive.google.com/open?id=1KsUF2SzkhvBOkHzbrXKj7H5KtN6Z3hRJ), [log](https://drive.google.com/open?id=155Ce-FmI6dmMgJnXdESVHx08unU3jWX2)]

        * FastAP, ResNet-50, M=96, Dim=512: [[model @ epoch 40](https://drive.google.com/open?id=1AblJelRHStBfWwmZeoRM8iEpNRNdOobn), [log](https://drive.google.com/open?id=1twswLE-j9kLxUsk5Ku7vWqBp0Sml65sG)]

        * FastAP, ResNet-50, M=256, dim=512: [[model @ epoch 30](https://drive.google.com/open?id=1MAimhKEyEfq2LDYnUaDburFH2YsUhrpA), [log](https://drive.google.com/open?id=1CtNk-wxSZToO703OvfK8ndFQzFnpOvVS)]

        

        (M=mini-batch size)

 * PyTorch code is a direct port from our MATLAB implementation. We haven't tried reproducing the paper results with our PyTorch code. **For reproducibility use the MATLAB version**. 

 * Note that the mini-batch sampling strategy must also be used alongside the FastAP loss for good results.

## Contact

For questions and comments, feel free to contact: kunhe@fb.com or fcakirs@gmail.com

## License

MIT