Semester project (Part 5): Final update

Due May 2, 2025 by 11:59pm Points 100

The aim of this last part of the semester project is to test your solutions on unknown data. By "unknown data" I mean a sequestered set of samples, not used (or seen) when you were designing your method in previous phases. Here is a general list of deliverables for this final part:

A report (no page limit, but try to be concise; 1000-2000 words should suffice) as a separate (from Parts 1, 2, 3 and 4) section of the readme.md in your GitHub that includes

Description of the test database you collected or downloaded: What is the size of the database? What is different when compared to the training and validation subsets? Why you believe these differences are sufficient to test your final programs? (10 points)
A classification accuracy achieved on the test set. Use the same metrics as in previous phase. (20 points)
Most of you should see worse results on the test set when compared to the results obtained on train/validation sets. You should not be worried about that, but please provide the reasons why your solution performs worse (with a few illustrations, such as pictures or videos, what went wrong). What improvements would you propose to lower the observed error rates? (20 points)
For teams: explain individual contributions of each team member (this is needed to have this assignment graded).

Imagine you want to present your final program to a friend or investor. And this presentation should be short and illustrative (that is: show what you did and how good it is). Prepare a short video, or a short slideshow with pictures, presenting how your final program works. Here are good examples prepared by former CV students. (20 points)
Push your final (to be graded) code along with instructions how to run it (either Walter or Louisa will do it to see how the final deliverable works on test data). Your program(s) should pick one example from the test set (please attach this sample to your code) and present the processing result. We should be able to run your programs without any edits -- please double check before submission that the code is complete. (30 points)