https://github.com/lromul/argus-bengali-ai

Kaggle | Solution for Bengali.AI Handwritten Grapheme Classification
https://github.com/lromul/argus-bengali-ai

Last synced: 17 days ago
JSON representation

Kaggle | Solution for Bengali.AI Handwritten Grapheme Classification

• Host: GitHub
• URL: https://github.com/lromul/argus-bengali-ai
• Owner: lRomul
• License: mit
• Created: 2020-01-05T23:05:03.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2022-11-22T07:56:18.000Z (almost 2 years ago)
• Last Synced: 2024-10-07T11:42:36.893Z (about 1 month ago)
• Language: Python
• Homepage: https://www.kaggle.com/c/bengaliai-cv19
• Size: 1.92 MB
• Stars: 3
• Watchers: 2
• Forks: 0
• Open Issues: 4
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Bengali.AI Handwritten Grapheme Classification

Source code of solution for [Bengali.AI Handwritten Grapheme Classification](https://www.kaggle.com/c/bengaliai-cv19) competition.

## Solution 

Key points: 

* Efficientnets

* CutMix, GridMask

* AdamW with cosine annealing

* EMA

## Quick setup and start 

### Requirements 

*  Nvidia drivers, CUDA >= 10.1, cuDNN >= 7

*  [Docker](https://www.docker.com/), [nvidia-docker](https://github.com/NVIDIA/nvidia-docker) 

The provided dockerfile is supplied to build image with cuda support and cudnn.

### Preparations 

* Clone the repo, build docker image. 

    ```bash

    git clone https://github.com/lRomul/argus-bengali-ai.git

    cd argus-bengali-ai

    make build

    ```

* Download and extract [dataset](https://www.kaggle.com/c/bengaliai-cv19/data) to `data` folder.

### Run

* Run docker container 

```bash

make

```

* Create a file with folds split

```bash

python make_folds.py

```

* Train model

```bash

python train.py --experiment train_001

```

* Predict test and make submission 

```bash

python kernel_predict.py --experiment train_001

```