-
Notifications
You must be signed in to change notification settings - Fork 0
Developer Notebook – Tim Watson
This notebook will contain partner meeting notes, scrum meeting notes, and project change notes for Tim Watson's portion of the Team 11 - CS499 project 'Restoration of Compressed Images' in conjunction with Lexmark.
-
GCCR Word Count for this Notebook: 1,677
-
Non-Notebook GCCR Word Count: ~3,100
-
Total GCCR Word Count: ~4,777
Due to some contact issues with Lexmark, Dr. Cooper's (Brian) emails were being sent to himself rather than me, our first meeting of the semester with our corporate contact at Lexmark came a little later than it should have.
The majority of the meeting was Brian and two of his colleagues giving us an overview of the project. They told us about the shortcomings of the teams in the previous semester. One of the team wasn't able to complete any functional code and the other saw very mitigating results. The first team only created a GUI and some small developer tools, while the other team focused more on the machine learning aspect of the project. Lexmark suggested that we could use the other teams' code, but after speaking with Mr. Chartos (Jeff), this was not within the bounds of the class's boundaries.
Brian gave us some tips on how to be a cohesive team and suggested that out team meet early and often to be effective and get our work done as soon as possible.
Below are some of the overview points that Brian and his team gave us:
1.)We will only be working with grey-scale images.
2.) Our goal is to use OCR (optical character recognition) to pull the text from the input scanned image. Our team developed some ideas later for how OCR could be used as an error metric for measuring our machine learning algorithm(s).
3.) Training data (and input images) will be provided to the team in the form of documents that are first printed, then scanned, and finally compressed. The compressed versions will be the provided training date and could be a smaller amount of data than desirable.
4.) The client (Brian) prefers concise documentation. Detailed but not overly complex or deep. Brian also prefers using a company messaging group in Microsoft Teams and thus meetings with the group will only be few and far between. This worries me, but I understand that Brian is busy with his work for Lexmark and the other CS499 team.
We did not meet with Jeff today as there was a lot of snow on the ground and school was cancelled for the day. It was the most snow I can remember for the past few days and it was honestly nice to have the day off.
Jeff asked each of us questions that followed the typical Scrum meeting format of: what did we work on last week, what are we working on this week, and if we had any concerns about the project going forward. Our team wasn't able to say too much as we were still waiting on Brian to set up the Microsoft Teams channel and provide us with the files to work with. Michael and I started to think about and consider options for our command line interface (CLI). We have all be learning Machine Learning and/or honing our Machine Learning skills so that we're ready when the time comes to work on our model.
Our team met for about an hour of the third floor of the Science Library. We met to plan, develop, and finalize the Project Plan assignment. I worked on document one and finished it somewhat easily as we had a clear vision of the work that needed to be done on each of the documents. Siyuan worked on the structure of the Requirements document while Nolan wrote the meat and potatoes of it to make up for his absence during the majority of the meeting, which was completely fine with everyone. Finally, Michael worked on on document three.
Jeff asked each of us questions that followed the typical Scrum meeting format of: what did we work on last week, what are we working on this week, and if we had any concerns about the project going forward. I'm beginning to see a motif to these meetings. Our team told Jeff about the slow start to communication with Brian and his team. I finally have access to the Teams channel, but none of my other team members have been able to access it at this point. Despite that our team has been working well together and we all seem to enjoy working together. I'm glad that despite Michael and I already being good friends hasn't hindered us from getting along with Nolan and Siyuan who are great partners so far.
The team met together at the William T. Young Library this afternoon, but I was not able to make it there due to illness. I was worried at first that I would not be able to meet the requirements of the Architecture assignment that the team left for me to do, but I found that I was able to work on it and finish my portion (the testing section, which was the most tricky part) more quickly and easily than I expected. Using our work on the Architecture document, the team made slides for the Architecture presentation on 3/4/19.
Our team met at the Study South to go over our Architecture presentation. Chelsea, who we worked with, was very nice and made good suggestions for improving our oration but wasn't very helpful with the presentation itself.
Jeff asked each of us the typical Scrum meeting questions: what did we work on last week, what are we working on this week, and if we had any concerns about the project going forward. We spoke about the architecture assignment and presentation and continued to express our worries about the lack of communication with Brian. I'm personally worried that our team will not be able to meet our goals without better direction.
Our team met to discuss our individual findings for both super-resolution and OCR algorithms. Nolan and Michael have been researching OCR open source softwares while Siyuan and I have been researching super resolution algorithms. We have all found some promising github repositories and will continue to work toward finalizing the open source software we will be working with.
Our team worked on the development of a coding plan for the project. Michael and I (90% Michael) worked on and finished the basic command line interface that the program will use as an interaction shell with the open source software we pipeline in.
Jeff asked each of us the typical Scrum meeting questions: what did we work on last week, what are we working on this week, and if we had any concerns about the project going forward. Jeff also briefed our team on the upcoming assignments and due dates. He suggested that we settle on the software that we would like to use soon so that we can formally start coding.
Jeff asked each of us the typical Scrum meeting questions: what did we work on last week, what are we working on this week, and if we had any concerns about the project going forward. Our team had very little so talk about as we had all been very busy with midterms and other class projects. We also voiced our continued concerns with communication with Lexmark.
Our team met to finalize our plan for the functional model of our project. We have made two significant decisions on the project as a whole. First we decided that we will not be working on OCR as our skills and time remaining are not conducive to working on it. Instead the team will be focusing its efforts on producing a functional super resolution machine learning model. Second we have selected Neural Enhance as our open source software of choice to work with. Michael has been working hard on figuring it out and preparing for training it on images of text rather than picture images.
Jeff asked each of us the typical Scrum meeting questions: what did we work on last week, what are we working on this week, and if we had any concerns about the project going forward. We told Jeff of our progress with Neural Enhance and of Michael's work to train it (it took three days!). There have been some issues with the training process, but overall the model seems to be working well with our pdf images. The only major defect comes with the color of the output image. For some reason the pages are slightly brown-red, but when run through a grey scaling program it looks very good. Near the same quality, except for some minor artifacting, and at 4 times the resolution.
Our next step is to work on getting the program to run on other machines that are more powerful. We are going to attempt to run it on my laptop which has a moderately powerful GPU to use and could extremely increase our efficiency.