Efficient Re-identification with Deep Learning clustering models
Motivation
Given streams of data not instantaneous to time of inference, for a traffic counting prompt, we can greatly reduce time and cost by applying clustering-based models instead of the traditional detection models.
Master Source
Learning to Cluster Faces on an Affinity Graph
Investigation Pipeline
As with any ML/DL models, the first step is one of the hardest steps to overcome in your reserach if not the very hardest. It did not help that the author of this code did not structure his repository intuitively, nor is it well-maintained. In short we have a few challenges to overcome:
- Input Preparation
- Training Inspection
- Output Interpretation
- Auxiliary Output Interpretation
- Application, Evaluation, and Decision