NVIDIA-ISAAC-ROS / NVIDIA-ISAAC-ROS/isaac_ros_object_detection
Add RF-DETR support (isaac_ros_rfdetr)
Nobody has claimed this yet.
- Dominant language
- C++
- Stars
- 201
- Forks
- 51
- PR merge metrics
- No merged PRs in 30d
Description
I'd like to add support for RF-DETR (Roboflow, Apache-2.0) to this repo as a new isaac_ros_rfdetr package, following the existing isaac_ros_rtdetr structure.
RF-DETR is currently one of the most accurate real-time detectors available. The Apache-2.0 variants reach 48.4 mAP at 2.3 ms (RF-DETR-N, 384x384) up to 56.5 mAP at 6.8 ms (RF-DETR-L, 704x704) on COCO, which beats the YOLO family at comparable latency.
Proposed scope
isaac_ros_rfdetr: decoder node (C++), launch fragments, visualizer script, POL test with a dummy ONNXisaac_ros_rfdetr_models_install: asset script to fetch weights and export ONNX -> TensorRT engine
I'd like to open a PR for this. Before I start, I just want to check that you're interested and that the scope above is the shape you'd want it in. If it looks good I'll begin work.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reading the existing isaac_ros_rtdetr structure and compare its decoder node, launch fragments, visualizer, and POL test with the proposed isaac_ros_rfdetr package. Review the model-install workflow for fetching weights and exporting ONNX to a TensorRT engine. Done means both proposed packages and their listed components are implemented and tested, with the scope confirmed by maintainers.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp, machine-learning
- Domain
- computer-vision, machine-learning, robotics
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 48/100