What did you personally accomplish this week on the project? Give files or photos that demonstrate your progress. Prove to the reader that you put sufficient effort into the project over the course of the week (12+ hours).
Overall a lot of progress was made this week in preparation for the final demo and presentation. We decided that rescoping our project away from natural scene braille detection and wearability was in our best interest. This is due to the complexity required for initial recognition.
In terms of the post processing section of the project, I was able to complete a significant amount of work in developing the spell checking algorithm even further. By taking in information about the projected confidence levels from kevin, I can check any number of characters throughout the string for corrections. After initial testing, this brings up the correctness to around 98-99% on single error words. In addition to confidence inclusions, I built a testing infrastructure that would run on large paragraphs of text and compare each word to a correct version after being run through the spell checker. For my algorithm, this gives me a clean interpretation of what is working best and what will work best after classification. I ran the test on both a static dictionary file with over 200,000 words, a text file containing sherlock holmes and little women with around 300,000 words, and a combination of both. The dictionary file provided the least accuracy due to the inability to produce a probability per word and limited size. The combination provided maximum confidence with very little efficiency loss. In addition, adding in the confidence matrix led to almost 2x speed ups for similar data sets. Of course these are simulated data sets so it will be interesting to run through the whole pipeline and see results.
In conclusion, we were also able to verify the sound capabilities of the jetson Nano and it was simple to connect to. This means that the full pipeline can perform on the jetson and we can run testing from pre-processing to post-processing.
Is your progress on schedule or behind? If you are behind, what actions will be taken to catch up to the project schedule?
We are currently still on track after rescoping to the stationary educational model. I think it would have been nice to have more data for the presentation but overall I am happy with where we are.
What deliverables do you hope to complete in the next week?
As we get into the final stretch, I think it is crucial to be able to add final testing metrics and data for the presentation and final demo. This will show that we thoroughly checked our options and effectively came to our final product with quantitative reasoning.