Skip to content

Latest commit

 

History

History
22 lines (17 loc) · 877 Bytes

README.md

File metadata and controls

22 lines (17 loc) · 877 Bytes

Bengali.AI Handwritten Grapheme Classification Pytorch

Original competition

Test recall calculated according to competition's evaluation metric

How to run training?

git clone https://github.com/meyson/bengaliai-cv19.git
cd bengaliai-cv19
kaggle competitions download -c bengaliai-cv19
unzip bengaliai-cv19.zip -d data
python3 create_folds.py
python3 create_train_images_png.py
chmod +x run.sh && ./run.sh
Name # layers # params Test recall
SqueezeNet - 1.3M ~0.95
EfficientNet B3 - 1.3M ~0.965