Skip to content

Releases: headertag/compare

v1.0.0

17 Nov 23:35
@headertag headertag
v1.0.0
d974de3

Choose a tag to compare

View all tags
v1.0.0 Latest
Latest

Release Notes v1.0.0

Compare v1.0.0 marks the first stable, major release of the ensemble object detection system. This release focuses on some minimum requirements for the web dashboard streaming output, robust multi-user alerting, and verified support for embedded NVIDIA hardware.

🚀 New Features

  • Live Background Removal: Added an optional feature to the web dashboard to remove video backgrounds in real-time using the briaai/RMBG-1.4 model. This can be toggled via the dashboard.remove_background setting in config.yaml.
  • Model Visualization: The dashboard now visually differentiates between the five ensemble models. Bounding boxes are color-coded (e.g., DETR is Red, YOLOS is Blue) so users can see exactly which model is detecting an object.
  • Video File Support: Provided instructions about running the system against raw video files (e.g., .mp4) instead of a live webcam feed by specifying the file path in the camera.index configuration.
  • NVIDIA Jetson Orin Nano Support: Comprehensive instructions and dependency adjustments (JetPack 6.2) have been added for running the ensemble on edge devices including the necessary startup script for containerized deployment.

🐛 Bug Fixes

  • Multiple Telegram Chat IDs: Fixed a bug where alerts were not broadcasting to all configured chat IDs. The alerting system now correctly iterates through the entire list in config.yaml.
  • Dashboard Latency: Applied tweaks to the dashboard streaming loop to improve frame update rates.

🔧 Improvements & Configuration

  • CPU Default: The default configuration has been switched to cpu to ensure the application runs out-of-the-box on standard hardware. CUDA/GPU support remains a toggle in config.yaml.
  • Refactored Codebase: Major cleanup of legacy code and improved cross-platform compatibility.

📦 Models Included

This release locks in the following ensemble configuration:

  • DETR (ResNet-50)
  • YOLOS (Tiny)
  • Faster R-CNN (ResNet-50)
  • RetinaNet
  • YOLOv5s

⬇️ Upgrading

If upgrading from a pre-release version, please ensure you update your config.yaml to include the new dashboard settings:

dashboard:
  remove_background: false
Assets 2
Loading