In this Era of Artificial Intelligence & Deep Learning, and with the aim of improving the customer's services in the supermarket industry, Picnic wants to ensure better customers services throughout labeling sent images according to the product that is in the picture.
What it does
As the challenge says, it is about labeling sent images from customers according to the product that is in the picture, this is a computer vision problems knows as Image Recognition that can be tackled using machine learning and deep learning model.
How I built it
Image Recognition tasks can be solved throughout Training a Convolutional Neural networks (CNN), by feeding it the dataset images provided by PicNic, for that I used Keras and FastAI as an abstraction for building and training my CNN's, also for loading the images and for doing Image Augmentation, of course, some other libraries like pandas, numpy and Matplolib used to do various task, and finally, to train massive CNN's, we need some GPU power, so I've Used Google Colab which provide free GPU power to train and build Neural networks.
Challenges I ran into
Each project has its own challenges, and the challenges I ran into was mainly to improve the performance with different techniques, like adding more data, doing data augmentation, delete misleading images, and also playing with the CNN hyperparameter like batch size, learning rate, the optimizer, and finally choosing the right pre-trained model and finetune it t, so I was testing the combination of all these things to get the best possible result.
Accomplishments that I'm proud of
The Main thing I'm proud of is the model I've built, and the accuracy I got, it was about 90% and this is very great cause it was really difficult (In my view) even for Humans.
What I learned
Always Hackathons make us learn new things especially when working with a team, I didn't get a chance to work in a team but eventually, I learned a lot of things from this real-life problem, I learned how to deploy a model, create and design different solutions, using others models, using a lot of frameworks and libraries and much more.
What's next for The PicNic Vision
Of course, the next thing for The PicNic Vision is to win this hackathon, next, is to use text recognition to handle things in images like the boxes of Prok, Beef & Lamb, Minced Meat and also poultry, since sometimes the customer send an empty box of them which our model can't decide based on the shape or the color of the box but instead, he must understand what has been written on the box, this improvement will lead to better accuracy and precision for the model.