Raw Sample Collector (RSC)
Raw Sample Collector
Raw Sample Collector (RSC) is a new feature introduced to improve detection performance in VIS. It allows the system to record raw[1] video frames from logged events. These raw frames are later used to evaluate and retrain the latest AI model, helping to identify missing detections or reduce false positives. By feeding corrected images back into the training process, RSC helps improve model accuracy right after next training.
RSC collects one thousand frames for each Vislet once activated. It is capable to run on multiple Vislets simultaneously, each operating independently and in parallel. Only frames corresponding to important logged violations are recorded. Images are temporarily stored and then automatically zipped once 1000 frames are collected. After zipping, the individual image files are deleted.
Housekeeping
Zipped RSC archives are cleaned up according to the standard housekeeping rules.
Purpose
The primary goal of RSC is to collect real world visual cases that current models might miss or misclassify. This targeted data collection enables faster and more relevant improvements to detection models.
Availability
- VIS version 26 SP1
How It Works
Activation
RSC is activated from the Visor interface using the RSC button in the camera tab.
Logging and Access
Each RSC session is logged with step no 126. Step no 126 is normally not visible in logs tab, it only appears if specifically filtered for it. A download link for the zip file is provided in the Visor logs tab.
- ↑ Raw means no detection, no region, no information section. Images are saved as collected from camera feed.