Choice 3: Sketches#
In this project, you will develop a content-based image retrieval system with PyTorch. You will focus on the case where the query is a hand-drawn sketch, and the goal is to find photos that match the category of the object depicted in the sketch. You will build a neural network that learns a similarity metric between sketches and photos, allowing the system to retrieve images from a photo gallery based on how well they visually correspond to the query sketch.
Difficulty |
Suggested Tutorials |
|---|---|
Moderate |
Requirements: This project can be completed on a standard laptop without a dedicated GPU. However, having access to a GPU will allow you to choose a larger pre-trained model (e.g., ResNet-50) and fine-tune it for potentially better performance.
Grading
The project will be graded based on the following criteria. Points for each activity are awarded based on quality and completeness (partial credit possible).
Activity |
Points (max) |
|---|---|
Build a baseline image retrieval system |
9 |
Properly evaluate its performance (precision, recall) |
3 |
Transfer learning (fine-tuning) |
3 |
Deployment (model made usable) |
2 |
Presentation (clarity & demo) |
3 |
Total |
0-20 |