Food recognition using streamlit with inception v3 backend
Deployed at Streamlit
Every one likes food! This deployment recognizes 11 different classes of food using a SOTA Inception V3 Transfer Learning.\n
-
The paper for Inception can be found here
-
The paper implementation using pytorch can be found here
-
Inception-v3 is a convolutional neural network architecture from the Inception family that makes several improvements including using
- Label Smoothing,
- Factorized 7 x 7 convolutions,\n and the use of an auxiliary classifer to propagate label information lower down the network (along with the use of batch normalization for layers in the sidehead).
-
Training on 16,600 images yielded 90% accuracy on train and 76% accuracy on validation. over 50 epochs!
-
This model is saved and used later
Data consists of 1.1GB of 16,600 images of different categories of food. the categories of food that can be classified are
- Bread
- Dairy Product
- Dessert
- Egg
- Fried Food
- Meat
- Noodles-pasta
- Rice
- Seafood
- Soup
- Vegetable-fruit
Dataset is obtained from kaggle