Cassava Leaf Disease Classification

Kaggle Competition / Report / Submission / Model Checkpoints / Training Data

Final Score

• Public Leaderboard: 0.9095 (Rank 4)
• Private Leaderboard: 0.9060 (Rank 2)

Installation

conda create --name SC4000 python=3.10
conda activate SC4000

git clone https://github.com/pufanyi/SC4000.git
cd SC4000
python -m pip install -e .

If you want to train the CropNet, you need to install tflite-model-maker-nightly manually:

python -m pip install --use-deprecated=legacy-resolver tflite-model-maker-nightly

Train

python -m sc4000.train.run --model=vit --subset=full

Evaluation

python -m sc4000.eval.evaluate --model=vit --subset=full

For the --model_args options, should input like this:

python -m sc4000.eval.evaluate --model=vit --model_args="pretrained=output/models/checkpoint-124" --subset=full

Citation

@misc{sc4000cassava2024,
    title        = {Cassava Leaf Disease Classification},
    url          = {https://github.com/pufanyi/SC4000},
    author       = {Anusha Agarwal*, Fanyi Pu*, Gordon Tian Xiao Chen*, Siyang Shao*, Yi Shan*},
    publisher    = {Zenodo},
    version      = {v1.0.0},
    month        = {December},
    year         = {2024}
}