- Emotion classification from videos
- Joachim Vanneste
- 2460800V
- Debasis Ganguly
- 1 hour Read the project guidance notes
- 0.5 hour Created GitLab repository
- 2 hours Read through Master dissertation provided by supervisor
- 1 hour Read through some articles referenced in Master dissertation
- 0.5 hour Set up GitHub Wiki for project meetings
- 1 hour Research into BeautifulSoup to crawl videos
- 3 hours Create basic crawler to crawl over classic tv commercial website
- 2 hours Fit crawler to website provided by supervisor
- 2 hours Finished web crawler
- 1 hour Research into how to download the videos from their links
- 3 hours Annotate research papers in Zotero
- 1 hour Bug fixes in crawlers
- 3 hours Find trade off between max ad size and number of ads (storage space full)
- 0.5 hour Gave videos id's rather than title as before
- 5 hours Downloaded videos - script was buggy so had to run it several times
- 2 hours Work on video annotation script
- 0.5 hour Research into OOD data and clustering
- 2 hours Annotated 100 of the downloaded advertising videos
- 0.5 hour Project meeting preparation
- 0.5 hour Project meeting
- 1 hour Active learning and bootstrapping research
- 2 hours Look through masters dissertation source code
- 2 hours Read through ALEX paper on activated learning
- 4 hours Prepare slides on AL and bootstrapping for meeting
- 0.5 hour Reach out to Jin for more info on source code
- 4 hours Try and get Jin's code to work - NOT GOING TO WORK
- 2 hours Project meeting preparation
- 0.5 hour Looked into who feasible it would be to redo Jin's project first
- 0.5 hour Meeting with supervisor
- 0.5 hour Write up project meeting minutes
- 6 hours Downloaded Pitts dataset and worked on Jin's code
- 4 hours Read and annotate Literature Survey on Active Learning
- 2 hours Play with parameters on model
- 2 hours Project meeting preparation
- 2 hours Research into AL query based frameworks
- 0.5 hour Project meeting
- 2 hours Found an optimal parameter configuration
- 3 hours Ran annotated videos through pre-trained models
- 5 hours Worked on uncertainty sampling function
- 2 hours Worked on uncertainty sampling function
- 1 hour Removed annotated videos from list of videos
- 2 hours Worked on uncertainty sampling function (completed)
- 1 hour Project meeting preparation
- 1 hour Project meeting
- 1 hour Split videos into test and train sets
- 3 hours Work on cross validation
- 3 hours Work on random sampling
- 0.5 hour Work on cross validation
- 2 hours Time distribution
- 0.5 hour Finish one fold cross validation
- 2 hours Finish and test random sampling
- 0.5 hour Create and update results.md
- 2 hours Update results and re-train models
- 1 hour Prepare for meeting
- 0.5 hour Meeting
- 2 hours Explainability in videos research
- 2 hours LIME and SHAP research
- 1 hour Plan for project
- 1 hour Prepare for meeting
- 0.5 hour Meeting
- 5 hours Work on finding n key frames in a given video
- 2 hours Additional LIME research
- 3 hours Mask out key frame segments and run on model
- 3 hours How to convert accuracy into similarity?
- 2 hours How do I extract weights
- 2 hours Meeting preparation
- 2 hours Run algorithm on several different videos
- 1 hour Meeting
- 5 hours Work on masking out pixels
- 1 hour More LIME research
- 2 hours Work on masking out pixels
- 2 hours Work on masking out pixels
- 2 hours Meeting preparation
- 4 hours Implement plotting (frames v frame importance)
- 2 hours Work on status report
- 2 hours Work on status report
- 2 hours Work on status report
- 0.5 hour Submit status report
- 0.5 hour Look over current project
- 0.5 hour Update supervisor on current problems
- 1 hour Think of solutions to problems
- 1 hour Start dissertation - background
- 2 hours Meeting preparation
- 1 hour Supervisor meeting
- 2 hours Work on explaining frames with masked out pixels
- 2 hours Try to use LIME to explain predictions
- 2 hours Test LIME with videos/images with a video model - does not work
- 2 hours Start implementing algorithm to perturbe local instance
- 5 hours Finish creating neighbourhood set for a given frame
- 0.5 hour Meeting preparation
- 0.5 hour Meeting
- 6 hours Start implementing parts of LIME algorithm (video explainer classes)
- 6 hours Decompile lime_image to see parameter shapes
- 4 hours Connected perturbed data and labels to LIME
- 2 hours Bug fixes
- 1 hour Try different regressors for surrogate model
- 2 hours Meeting preparation
- 0.5 hour Meeting
- 4 hours Start creating my own regression model with Keras
- 1 hour Supervisor meeting
- 5 hours Feature extraction to use ridge regression (average of frames)
- 3 hours Bug fixes
- 2 hours Show mask
- 4 hours Implemented new feature extraction method
- 2 hours Hyper-parameter tuning
- 1 hour Meeting preparation
- 1 hour Meeting
- 1 hour Meeting write-up
- 5 hours Dissertation work
- 2 hours Cherry picking examples
- 2 hours Start basic Flask app
- 2 hours Dissertation work
- 3 hours Work on app - js and css implemented
- 3.5 hours Work on app - home page work
- 2 hours Dissertation work
- 1 hour Dissertation work
- 2 hours Work on app - nicer css added
- 3.5 hours Show specific outputs
- 0.5 hours Meeting preparation
- 0.5 hours Meeting
- 4 hours Start demo paper
- 0.5 hours Create system design image
- 3 hours Dissertation work
- 3 hours Expand system to show explanations for funny
- 3 hours Work on demo paper
- 6 hours Work on demo paper
- 1 hour Meeting
- 3 hours Work on demo paper
- 2 hours Work on dissertation
- 3 hours Finish demo paper
- 2 hours Work on dissertation
- 1 hour Meeting preparation
- 0.5 hours Meeting
- 1 hour Meeting write-up
- 3 hours Evaluation reseach
- 2 hours Work on dissertation - background section
- 4 hours Implement SHAP
- 4 hours Fidelity and stability research
- 1 hour Meeting preparation
- 0.5 hours Meeting
- 1 hour Meeting write-up
- 3 hours Work on fideleity and stability scores
- 2 hours Work on dissertation - background section
- 3 hours New examples
- 2 hours Work on dissertation - implementation section
- 2 hours Work on dissertation - implementation section
- 6 hours Work on dissertation - implementationa and design sections
- 2 hours Work on dissertation - design section
- 4 hours Fidelity and stability calculations
- 1 hour Meeting preparation
- 0.5 hours Meeting
- 4 hours Work on dissertation - evaluation section
- 8 hours Work on dissertation - introduction and implemntation sections
- 8 hours Work on dissertation - evaluation section
- 2 hours Video presentation slides
- 6 hours Work on dissertation - evaluation and implementation sections
- 8 hours Work on dissertation - design and implementation sections
- 2 hours Work on dissertation - Implementation sections
- 2 hours Work on dissertation - Implementation sections
- 2 hours Work on dissertation - Look over supervisor comments
- 8 hours Work on dissertation - background and introduction sections
- 2 hours Work on dissertation - conclusion section
- 6 hours Work on dissertation - bit of everything
- 6 hours Work on video presentation
- 2 hours Finishing touches
- 0.5 hours Sumbit :)