Data-Centric AI Competition

Getting Started

Perform augmentation:

poetry run python -m augment -s experiments/original -t experiments/manual-clean-aug-quant -n 0.95

Find and eliminate duplicate images.
Create images with vertical and horizontal lines (see examples in train/i and train/ii)
Make digits centred into the frame.
Add noise (salt and peper) to the bottom of images
- See train/iii/b0607164-ce5d-11eb-b317-38f9d35ea60f
- See train/v/af3c6e78-ce5d-11eb-b317-38f9d35ea60f
Add blob on corner of the image (see train/iv/ae7cec06-ce5d-11eb-b317-38f9d35ea60f)
Determine if edge pixels are part of the main image.

Name		Name	Last commit message	Last commit date
Latest commit History 114 Commits
assets		assets
data		data
experiments		experiments
label_book		label_book
notebooks		notebooks
.gitignore		.gitignore
.python-version		.python-version
README.md		README.md
augment.py		augment.py
augment_v2.py		augment_v2.py
augment_v3.py		augment_v3.py
convert.py		convert.py
copy_cleaned.py		copy_cleaned.py
dash_app.py		dash_app.py
detect_duplicates.py		detect_duplicates.py
evaluate.py		evaluate.py
extract_features.py		extract_features.py
fine_tune.py		fine_tune.py
model.py		model.py
original_train.py		original_train.py
poetry.lock		poetry.lock
predict.py		predict.py
pyproject.toml		pyproject.toml
score.py		score.py
streamlit_app.py		streamlit_app.py
train.py		train.py
utils.py		utils.py