Week 5

Oct 10/24/2022

1. Link for using AI and running python code on Raspberry PI embulators was sent in an email.
2. Had meeting with Vannary to discuss the previous runs made over the weekend.
   1. Asked her to put Collimator on hold.
   2. Discussed ways to generate various splits of the training dataset.
3. Vannary plotted charts with distributions of different labels in each video.
4. Nuances of framed being labelled incorrectly came into picture.
5. This is documented by Vannary.

Oct 10/25/2022

1. Fixed the bug with the huggingface deployment.
2. Went out for lunch with Sudhir, Jake and Dr. Dion
3. Shifted focus to the streamlit app for the SD1 demo on Thursday.
4. Clarified Dick's concern on data not having the desired elevation columns.
   1. Update: Dick was looking at the files sent in the previous email.
5. Had a meeting with Vannary to discuss issue with the weights and biases callback.

Oct 10/26/2022

1. Started working on the Streamlit app for the SD1 demo tomorrow.
2. Had a meeting with Sudhir to revise the design with an image carousel.
3. Meeting with Pavan:
4. Meeting with Vannary and Sudhir:
   1. Interpretation of results for the Learner object

Oct 10/27/2022

1. Sent the slide deck to Sudhir. Pushed remaining changes on the app to the repo.
2. Sent out emails regarding automated image annotation and dealing with class imbalace
3. Did some research on using SMOTE on image datasets.
   1. Got one rudimentary implementation ready.
4. End of the day, almost got the image carousel for un-blurred images ready.
   1. Needs just the look-up functionality to be implemented

Oct 10/28/2022

1. Finished the app with the functionality to show images from any csv.
2. Based on the selected csv, got an updated carousel at the bottom for un-blurred images starting from the selected image.
3. Had a meeting with Pavan and Vannary, discussed a lot of things pertaining to:
   1. Data organization
   2. Trained models
   3. Nuances in the frame capturing scripts and spatial correlation
      1. Could be explored further, the algorithm used for calculating the spatial correlation.
      2. Link from Data Science StackExchange
   4. Meeting notes were captured by Deven, Vannary and Pavan and were sent out. Email has the subject, Collective meeting notes on CCTV
4. Meeting with Vannary
   1. Tried to plot confusion matrix for our multi-label classification model.
   2. Confusion matrix did not turn out as expected.
   3. A question was asked on the forums, and the response after the end of the day.
   4. Plotting confusion matrix for multi-label classification models is not yet supported.
5. Meeting with Pavan
   1. Had a discussion on code organization
   2. Making modular functions and potentially making an image-utils and opencv python package of GQC using nbdev.
      1. For modular use and reusability of same code.
      2. Needs to be discussed with Sudhir
6. IDEF0 diagrams for nuances of Jupyter notebooks and Google Drive was sent out.

Oct 10/31/2022

1. Morning meeting with Jake and Sudhir. Points discussed and recap
   1. Meeting recap with SD1
      1. Solo
      2. Truck
      3. Sent out an email to Roboflow guy for getting started with image annotation.
   2. Typo to fix in Vannary's update. Have a meeting with Vannary to fix this.
   3. Definition of precision, recall and true positive in case of multi-label classification problem from a book.
      1. Custom fastai implementation of MultiLabelClassificationInterpreter
   4. Python package for image-utils. Package is named as gqcv.