Vision Models
Collection
Vision models for object detection and image recognition. Includes RT-DETR series (v1/v2/v3/v4) and other computer vision models. • 2 items • Updated
🚀 RKNN Object Detection Models
ONNX + RKNN models for object detection, optimized for Rockchip RK3588 NPU deployment via rknn_model_zoo.
📦 Models
ONNX (Universal Format)
| Model | File | Architecture | Size | Source |
|---|---|---|---|---|
| YOLOX-S | yolox/yolox_s.onnx |
CNN | ~24 MB | examples/yolox |
| YOLOX-M | yolox/yolox_m.onnx |
CNN | ~68 MB | examples/yolox |
| PP-YOLOE-S | ppyoloe/ppyoloe_s.onnx |
CNN | ~24 MB | examples/ppyoloe |
| PP-YOLOE-M | ppyoloe/ppyoloe_m.onnx |
CNN | ~68 MB | examples/ppyoloe |
RKNN (RK3588 NPU Format, INT8 Quantized)
| Model | File | Quantization | Size | Tool |
|---|---|---|---|---|
| YOLOX-S | yolox/yolox_s.rknn |
INT8 | ~11 MB | rknn-toolkit2 2.3.2 |
| YOLOX-M | yolox/yolox_m.rknn |
INT8 | ~27 MB | rknn-toolkit2 2.3.2 |
| PP-YOLOE-S | ppyoloe/ppyoloe_s.rknn |
INT8 | ~9.3 MB | rknn-toolkit2 2.3.2 |
| PP-YOLOE-M | ppyoloe/ppyoloe_m.rknn |
INT8 | ~25 MB | rknn-toolkit2 2.3.2 |
RKNN files are converted from ONNX with INT8 quantization. Input/output dtype changed from float32 to int8 for NPU performance.
Original Download URLs
| Model | ONNX |
|---|---|
| YOLOX-S | yolox_s.onnx |
| YOLOX-M | yolox_m.onnx |
| PP-YOLOE-S | ppyoloe_s.onnx |
| PP-YOLOE-M | ppyoloe_m.onnx |
📥 Download
from huggingface_hub import hf_hub_download
# Download ONNX (universal)
hf_hub_download(repo_id="eebyp/RKNN", filename="yolox/yolox_s.onnx", local_dir="./model")
# Download RKNN (RK3588 NPU, INT8)
hf_hub_download(repo_id="eebyp/RKNN", filename="yolox/yolox_s.rknn", local_dir="./model")
🔗 Source
- Repository: airockchip/rknn_model_zoo
- Platform: RK3562 / RK3566 / RK3568 / RK3576 / RK3588
- Conversion: rknn-toolkit2 2.3.2, target_platform=rk3588, INT8 quantization
📄 License
Apache 2.0
Note: This repository is hosted for personal research purposes only. For production use, please refer to the official repository above.