kganeshchandan / Multi-Handwritten-digit-recognition-CNN

Multi-Handwritten-digit-recognition-Model

The model was constructed by using pytorch libraries and builtin dataprocessing methods in openCV

Instructions to run the Model

• clone this repository

• Install few packages like torch, tqdm, seaborn etc

• If facing issues with dependecies then refer to the conda environment yml file to set up the environment required to run th files

• Run the Simple-MNIST-CNN.ipynb notebook to create a CNN network trained on MNIST dataset and save the model

• To test the model on the new test datasetrun the Testing-MNIST-Model.ipynb

• The pretrained models have been saved in MODELS directory.

• The best model so far is alredy imported in the Testing-MNIST-notebook.ipynb by default

loss plot

Prediction vs true values on the new dataset

The best performance so far

76.38 %