https://github.com/filapro/visda2019

MixMatch Domain Adaptation: Prize-winning solution for both tracks of VisDA 2019 challenge
https://github.com/filapro/visda2019

domain-adaptation efficient-net mixmatch tensorflow visda

Last synced: 9 days ago
JSON representation

MixMatch Domain Adaptation: Prize-winning solution for both tracks of VisDA 2019 challenge

• Host: GitHub
• URL: https://github.com/filapro/visda2019
• Owner: filaPro
• License: other
• Created: 2019-08-02T23:08:39.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2022-11-21T21:17:22.000Z (about 2 years ago)
• Last Synced: 2023-03-04T22:58:18.773Z (almost 2 years ago)
• Topics: domain-adaptation, efficient-net, mixmatch, tensorflow, visda
• Language: Python
• Homepage:
• Size: 112 KB
• Stars: 22
• Watchers: 3
• Forks: 2
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
This codebase contains the implementation of our (with [@denemmy](https://github.com/denemmy))

solution for [VisDA 2019](http://ai.bu.edu/visda-2019) challenge. 

Our team got **2nd** place on final leaderboard of 

[multi-source](https://competitions.codalab.org/competitions/20256#results) track (with accuracy: .716),

and **3rd** place of [semi-supervised](https://competitions.codalab.org/competitions/20257#results) (with accuracy: .713).

This solution heavily borrows ideas from 

MixMatch ([arxiv](https://arxiv.org/abs/1905.02249), [github](https://github.com/google-research/mixmatch)) 

and EfficientNet ([arxiv](https://arxiv.org/abs/1905.11946), [github](https://github.com/qubvel/efficientnet)). 

The technical report is published on [arxiv](https://arxiv.org/abs/1910.03903).

#### Installation

Just clone this repo, update `PYTHONPATH` and install `requirements.txt` throw `pip`.

The code was tested on `ubuntu 16.04` with `python 3.6`, `cuda 10.0`, `cudnn 7.5`.

You may also need `wget` and `unzip` packages to download data.

#### Data preparation 

Download and convert images to `.tfrecords`:

```

python scripts/download.py

python scripts/convert_to_tfrecords.py

```

The resulting structure of data directory is shown in [docs/structure.md](docs/structure.md).

#### Training example

```

python runners/source_semi_supervised.py

```

The growth of accuracy on sketch domain will be displayed at `stdout` and in log file.

The arguments of all scripts are listed in [docs/arguments.md](docs/arguments.md).

#### Achieving leaderboard accuracy

Follow the instructions in [docs/solution.md](docs/solution.md).