# ICV Project Proposal ## Artemii Bykov, Maxim Salo B17-DS-01 ###### tags: `Introduction to Computer Vision`, `Multiclass Classification`, `Supervised learning`, `Food` 1. **Project title:** Multiclass Food Classification 2. **Project idea:** The model should recognise food by image. It is very useful and can easily applicable to different areas: markets, agricultural industry and services for food delivery (with robots) 3. **Technique/Method:** * 1st approach: * As a base we are going to use pre-trained **InceptionV3** - a convolutional neural network for assisting in image analysis and object detection * Pretrained means that it already has **learned weights** * We want to add a few layers on the top to tune the NN for our dataset * It will help us in faster convergence and saved time (Moreover, we have a huge dataset) * 2nd approach: * We want to try Wide-Slice Residual Networks with Slice Branch Network (you can find paper in the reference part) 4. **Dataset Explanation and [link](https://www.kaggle.com/dansbecker/food-101):** * This dataset consists of 101 food categories, with 101'000 images. For each class, 250 manually reviewed test images are provided as well as 750 training images. 5. **Timeline:** * *Artemii Bykov*: Parse dataset and perform data augmentation * *Maxim Salo*: Read the docs for InceptionV3 and understand how to finetune the model 6. References * https://www.kaggle.com/theimgclist/multiclass-food-classification-using-tensorflow/data * https://www.tensorflow.org/datasets/catalog/food101 * https://github.com/DucLeTrong/food101-classification * https://github.com/aurotripathy/food-classify * https://arxiv.org/pdf/1612.06543.pdf