What did you personally accomplish this week on the project?
- Further Yolo Testing
- This week I added in various Scripts and functions aimed at robustly testing each YOLO test configuration. While there was iterative testing before, now it was far more clear to see the data from each run and what commands led to it. In addition I added in functionality to log if an image was used in training/val/testing for each run so I could replicate runs exactly.
- In addition I completed additional labelling based off of previous results to ensure a more robust training dataset with more items and categories. At this point MVP has been achieved and I’m just expanding on proven ways to talk about dataset.
- I have looked into further ways to expand accuracy apart from YOLO – most notably grounded dino which is an openset detector which can look into finding an item if YOLO doesn’t know of a certain category. It even has relatively good accuracy compared to YOLO on known categories so I am currently testing a combined program for accuracy and will compare it to YOLO’s raw accuracy
- Ethics Assignment
- I spent a non-zero time completing the ethics assignment though this has no direct relation to the rest of the technical work this week.
Is your progress on schedule or behind? If you are behind, what actions will be
taken to catch up to the project schedule?
Given that the MVP has been achieved and a way to train YOLO has been proven to some extent I would say things are on schedule. At this point engineering time is going into ensuring the YOLO model gets even more robust and our solution scales.
What deliverables do you hope to complete in the next week?
Next week I hope to have the results of the grounded dino integration and hopefully get on with audio processing if the former goes well. And of course I will keep developing the test framework with more comprehensive traces including possibly a full integration test.
0 Comments