Came across a youtube video where Daniel replicated the Airbnb amenity detection task from scratch with data gathering to preprocessing to training an existing model to deploy and creating an App for inference. Just the first few minutes and I knew I had to try this out. It covers everything one needs to learn about how Deep Learning works and applying it to solve a real-world problem.

To explain the task is detecting Amenities from a photo uploaded of a room, Airbnb is a platform where you can rent a house or apartment. So what is the business use case here, historically a user uploads photos of the property and adds some details as well as the amenities that come with it, Amenity is something like luxury or that can be used, like a couch or cabinet, etc. So instead of the user adding the amenities why not detect them ourselves from the images. That's where DeepL comes in in form of Object Detection. So let's give it a try Shal we !!

Criteria

So what defines success for us or when can we say we have achieved something.

• [ ] It should be an app where I can upload a photo and it shows me all the objects it detected.
• [ ] Try to beat or equal Airbnb's MVP of 50% mAP using Google's AutoML .
• [ ] Model should be downloable which is not the case for AutoML.
• [ ] Training and inference has minimal costs.

Resources

Week 1:

TODOs for this week:

• [x] Gathering all related resources
• [x] Read the original post from airbnb to understand how they approached the problem
• [x] Read post from Daniel and studied how he is using Detectron2 for his model