RTMDet: An Empirical Study of Designing Real-Time Object Detectors
Paper • 2212.07784 • Published • 1
RTMDet is an efficient real-time object detector that exceeds the YOLO series, featuring a model architecture with large-kernel depth-wise convolutions and soft labels in dynamic label assignment. It is easily extensible for instance segmentation and rotated object detection tasks.
Original paper: RTMDet: An Empirical Study of Designing Real-Time Object Detectors
RTMDet-Nano
This model uses the RTMDet-Nano variant trained specifically for person detection. It is designed to work with RTMPose in a two-stage pipeline for real-time human pose estimation: RTMDet first detects persons in the image, then RTMPose estimates the keypoints for each detected person.
Model Configuration:
- Reference implementation: Official MMDetection RTMDet models
- Original Weight: rtmdet_nano_320-8xb32_coco-person
- Resolution: 3x320x320
- Support Cooper version:
- Cooper SDK: [2.5.3.1]
- Cooper Foundry: [2.2]
| Model | Device | compression | Model Link |
|---|---|---|---|
| RTMDet-nano | N1-655 | Amba_optimized | Model_Link |
| RTMDet-nano | N1-655 | Activation_fp16 | Model_Link |
| RTMDet-nano | CV7 | Amba_optimized | Model_Link |
| RTMDet-nano | CV7 | Activation_fp16 | Model_Link |
| RTMDet-nano | CV72 | Amba_optimized | Model_Link |
| RTMDet-nano | CV72 | Activation_fp16 | Model_Link |
| RTMDet-nano | CV75 | Amba_optimized | Model_Link |
| RTMDet-nano | CV75 | Activation_fp16 | Model_Link |
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support