Cassava Leaf Disease Classification is a Kaggle's Research Code Competition where you have to identify the type of disease present on a Cassava Leaf image.
The notebook is available on Kaggle to work in the same environment where this notebook was created i.e. use the same version packages used, etc...
If you are interested in the model the you can find that in the output section of the notebook.
Since in this competition you cannot use TPU so the model is trained using TPU and then a separate notebook is created which uses GPU and that is submitted. Here old submission notebook and new submission notebook are there. Reason for 2 notebooks is because I had some issues in loading the model dataset in old notebook, so I created a new one.
The model trained for this project with and without XLA and Mixed Precision is saved on Kaggle as dataset and can be found here
In this competition, the dataset has 21,367 labeled images collected during a regular survey in Uganda. Most images were crowdsourced from farmers taking photos of their gardens, and annotated by experts at the National Crops Resources Research Institute (NaCRRI) in collaboration with the AI lab at Makerere University, Kampala. This is in a format that most realistically represents what farmers would need to diagnose in real life.
Your task is to classify each cassava image into four disease categories or a fifth category indicating a healthy leaf. With your help, farmers may be able to quickly identify diseased plants, potentially saving their crops before they inflict irreparable damage.
- For augmentation Albumentations package is used
- Also for augmentation CutMix, MixUp and FMix techinques are uses
- Custom learning rate scheduler
- Cross validation and Ensemble method
- Trained for 100 epochs (last model)
- Use of tf-records and tensorflow.data.Dataset
Labels names and count plot
Sample images
CutMix
MixUp
FMix
Results (for model trained using TPU, with XLA and Mixed Precision)
Competition score (Submission results for the model trained using TPU, with XLA and Mixed Precision, whose accuracy, loss and confusion matrix plotted above)
There are multiple issues that were faced during the process and many of theme still exists. All of these existing issues are summarized in the respective notebooks(.ipynb) where they appear.
NOTE: All the notebooks are completely functional and there are no issues while running, the issues are I talked about are not used for training the model but code and issue details can be found in respective notebooks.