cassava-classifier 0.0.3

Cassava leaf disease classification using Deep neural network in Pytorch

Cassava leaf disease classification

The idea of this project is to build an image classifier to find out healthy and diseased cassava leaves.

There are 4 different classes of leaf diseases namely - Cassava Bacterial Blight (CBB),Cassava Brown Streak Disease (CBSD),Cassava Green Mottle (CGM) and Cassava Mosaic Disease (CMD) .

• Training data can be found on the Kaggle competition page

• Streamlit app code can be found here.

Github discussion forum

Installation

pip install cassava-classifier

Inference example

import PIL import Image
from cassava.pretrained import get_model

image = Image.open("<insert your image path here>")

# Use cassava.list_models() to list of available trained models
model = get_model(name:str)
model.predict_as_json(image: np.array)
>> {"class_name":str, "confidence": np.float}

Try out the inference code on either google colab or kaggle.

Training pipeline

1.Model Architecture - Efficeientnet-B4 , Noisy Weights
2.Image Size         - 512
3.Optimizer          - Adam
4.Scheduler          - GradualWarumUpScheduler
5.Loss               - Focal Cosine Loss
6.Augmentations      - Hard Augmentations
7.Epochs             - 10
8.Early Stopping     - No
9.Mixed Precision    - Yes

Blog

[Medium link]

Acknowledgements

We would like to thank Kaggle community as a whole for providing an avenue to learn and discuss latest data science/machine learning advancements but a hat tip to whose code was used / who inspired us.

1. Teranus
2. Nakama