The project task is to classify the fruits in supermarkets based on deep CNN model. But the machines in supermarkets only have CPU. So my team select mobilenet as the basic NN model and train it to classify different types of fruits.
This project uses pytorch to implement Mobilenet with Jupyter Notebook. The main dataset consists of imagenet data and custom data.
Hope this could help you to enjoy deep learning.
Version 1
In version 1, we conduct two experiments on fruits classification.
First, we use watermelon and kiwi pictures from imagenet to fune-tune our trained mobilenet model. Then we validate it by our own collect data.
The own collect data size:
watermelon: 18 , kiwi: 16
And we can achieve 100% validation accuracy.
Then, we add three other fruits and repeat the experiment.
Training Dataset
We select Tiny Imagenet to train the mobilenet model.