Task A: Build s server that can interactively perform model inference.
- I have developed the computation module for parallel processing. This module using ZeroMQ for messaging between processes. VideoProcessingServer and TrackProcessingServer implemented in the context of this module.
- The application is developed using PySide2 library and it supports user interactions like video seeking.
- Heatmaps are processed to extract peaks using local peak finding and displayed at the user interface.
Task B: Implement a multi-object tracker to assign the predicted centroids to the correct animal over time.
- Kalman Filter is used to track detections and these detections assigned to the objects over time with label and trace information.
- VideoProcessing (inference + peak finding) and TrackProcessing statistics calculated and displayed at the status bar in realtime.
- Tracking algorithm implemented in parallel.
Clone the repository.
- git clone https://github.com/serkanishchi/zerosleap.git
- cd zerosleap
Create Virtual Env.
- python3.8 -m venv env
Activate.
- source env/bin/activate
Install requirements.
- pip install -r requirements.txt
Setup package.
- python setup.py install
Download Pretrained Model.
- https://www.dropbox.com/s/kuyqdopree4fh0r/best_model.h5?dl=1
- Put the model under the "zerosleap/resources" folder with name of keras_model.h5
Run the application.
- python zerosleap/main.py