All notable changes to this project will be documented in this file.
-
Added face recognization and face boundary detection @nelson
- Using transfer learning and triplet loss on the embeddings to classify faces with few examples
- Experimented other features and other models for the same, and using HOG features, speeds up real-time
-
Added video to training picture script @eric
- Frame-by-Frame splitting of a video, to store the train image for a person
- Depth estimating using face size
- Created scripts to simplify running and creating a model
-
Visualizing the marauders map as a web-app @supriya
- Created layout in p5js
- updated requirements to ease setup
-
Setting up kinect@ashwin
- Connected kinect and visualizing depth map
- experimented with model types
- Testing and video creation
-
Project manager @talia
- worked with ashwin on enter name GUI pop up
- working on commands for taking pictures of user (telling them to move face)
- Found and troubleshot bug related to use on Windows machines
- Tested installation process on fresh macOS environments
- Supervised other team members
-
Results :pictures
-
Youtube :
- Refactored into classes@ashwin
- Kinect connected to live feed with depth@ashwin and @talia
- Visualizing 2d map in pygame @ashwin : not used for final iteration
- Made heroku website and added websocket code to the recognizer. just_runC.py runs without kinect code @eric
- Ported and setup processing visualizer on web-app@supriya
- Training module for each user. Users able to add their faces @nelson
- Trying out kinect libraries and experimenting with various version of depth map @ashwin](https://github.com/ashwinroot) and @talia
- Using angles,scales for relative distance @nelson and @eric
- Fixed the web app and visualizer @supriya
- Using camera.json config for multiple camera support @eric
- Removed kinect support as it was buggy
- Centroid tracking: tracking if a person enters a scene @ashwin
- Working on google drive interaction for a distributed system, for communicating a newly trained person @nelson
- Redid website to support multiple cameras in multiple canvas @supriya
- Fixed the data pipeline. Scaled-Position and person name calculation was moved to client side.@eric
- tkinter prompt to add a new user @talia
- Adding thresholds for distances and feet angle calculation@eric
- Redid the website to visualize multi-camera data in a single map representing the expo room along with real-time scaling@supriya
- Directions to train a new face on the frame. @talia
- Optimization and increasing the confidence in the classification of the person @ashwin
- Poster : all
- Video : @nelson