eee37
commented
7 years ago Things To Like: I like the complexity of the project. It sounds like a challenging project especially since your input will be images. In class we haven’t seen an example of this yet. I like the diversity in the type of features your project will incorporate (i.e images, text, etc.) You managed to circumvent the “messy” aspect of data science by finding a clean dataset. This should give your team more time to focus on the complexities of your project.
Areas of Improvement: It is still a little unclear to me how the business you are submitting the proposal to could benefit from the result of your study (What type of business is it?) Maybe providing a concrete example would of helped. In your proposal you hinted that multiple similar models already exist. So what new ideas will you be incorporating into your model? How will your model be different from existing ones? You did not really address the size of the dataset. Does that one dataset provide enough data? Do you have enough data to perform your analysis?
Thank you for the project submission! It was a pleasure to read through your proposal – it seems to be very well thought out and enthusiastically presented. I am excited to follow the progress of this assignment and see your results.
More specifically, I like the choice of dataset. Google Maps accumulates such an enormous amount of data which offers immense opportunity to draw conclusions from the images taken from streets and cities worldwide. Also, I like the application of reading traffic and detour signs. This is a smart application beyond the restaurant and business data that Google Maps already utilizes, and would have clear benefits for rerouting traffic. Lastly, the idea of drawing from existing character recognition techniques seems wise. Character detection seems to be a complex and growing research field, and approaching it with the help of existing techniques can facilitate the process.
I am slightly unsure with the exact idea of the output. It does not seem clear what exactly you want your output to be, as you mention a database of businesses and also mention mapping traffic and detour signs, so the specific goal does not seem initially clear. Also, I think you may be underestimating the “messiness” of the data you will be analyzing. Although, yes, it may be unlikely that pixels are missing, I think the threat of having unclear or distorted images may be very real. Lastly, the goal of improving existing character recognition classifiers seems to be a daunting one. Google Maps and others I’m sure already utilize robust image processors, so the capacity to improve on that may be limited. However, my knowledge of character recognition in natural images is very limited and so I likely do not know the scope of the room for improvement in that type of technology.
This project proposal is well constructed and the applications of the project are immediately tangible and applicable. Your sincerity of commitment to the project is promising, and I would be excited to see your results.