This project is submitted to AI Saturday (AI6), Kattankulathur, catalysed by Nurture.AI.
This project uses ResNet-50 to classify image of common Indian food. With limitation of access to GPU and no exiting accumulated dataset of India food over the internet. It implements transfer learning by fine tuning the last 10 layer of the ResNet-50 pretrained model on imagenet dataset. As an application of this indian food image classification model, it has simple Django web application interface to generate ingredients of the predicted food image by the fine-tuned ResNet-50 model.
With the help of transfer learning, model achieved good accuracy on only 200 images from each class.
# classes | 7 |
# image from each class used | 200 (due to limited RAM space) |
Training Accuracy | 96.79 % |
Test Accuracy | 91.43 % |
Here are pictures of the project demo.
$ pip install -r requirement.txt
$ cd foodImageClassifier
$ python manage.py runserver
Download the pre-trained model from here
training and evaluating model are done in this notebook. Download the dataset from here