Indian-Currency-Recognition

A project for visually impaired people to recognise the bills of Indian National Rupee using Deep Learning technology.

Dataset:

currency\
  ten\
  twenty\
  fifty\
  hundrad\
  fivehundrad\
  thousand\

dataset can be downloaded from here

Project sturcture based on CS230 Deep Learning

data/
    train/
    dev/
    test/
experiments/
    base_model/
        params.json
        ...
    learning_rate/
        lr_0.1/
            params.json
        lr_0.01/
            params.json
    batch_norm/
        params.json
model/
    input_fn.py
    model_fn.py
    utils.py
    training.py
    evaluation.py
build_dataset.py
train.py
search_hyperparams.py
synthesize_results.py
evaluate.py

Here is each file or directory’s purpose:

• data/: contain all the data of the project, with an explicit train/dev/test split.
• experiments: contains the different experiments (like base_model, learning_rate, batch_norm). Each directory (experiment) after training will contain multiple things:
  • params.json: the list of hyperparameters, in json format
  • train.log: the training log (everything print to the console)
  • train_summaries: train summaries for TensorBoard
  • eval_summaries: eval summaries for TensorBoard
  • last_weights: weights saved from the 5 last epoches
  • best_weights: best wieghts (based on dev accuracy)
• model/: module defining the model and function used in train or eval.
  • input_fn.py: where we define the input data pipeline
  • model_fn.py: creates the deep learning model
  • utils.py: utilitu function for handling hyperparams / logging
  • training.py: utility functions to train a model
  • evaluation.py: utility functions to evaluate a model
• build_dataset.py: creates or trainsforms the dataset, built the split into train/dev/test in reproducible way.
• train.py: train the model on input data, and evaluate each epoch on the dev set
• search_hyperparams.py: run train.py multiple times with different hyperparameters
• synthesize_result.py: expolre different experiments in a directory and display a nice table of the results
• evaluate.py: evaluate the model on the test set (should be run once at the end of the project
• app/: django project for interface

Django Interface:

pip install -r requirement.txt
cd app
python manage.py runserver

• There is module called app/classifier/predict.py which builds the tensorflow CNN model based on your experments, preprocess the image and load the all weights saved in the checkpoints. To do that you need set model_dir path in app/classifier.py/view.py.
• Model builder code is same as the model/

Refrences

• How to use transfer learning and fine-tuning in Keras and Tensorflow to build an image recognition system and classify (almost) any object
• Transfer Learning: retraining Inception V3 for custom image classification
• How to Retrain an Image Classifier for New Categories