Awesome
Jetson-nano-benchmark
Backbone | Network | Resolution | Mode | Inference Time | Framework | Task |
---|---|---|---|---|---|---|
pelee | ssd | 304 | float32 | 53 ms | TensorRT-Pelee | Detection |
pelee | ssd | 304 | float32 | 89 ms | TensorRT-Pelee | Detection + Segmentation |
mobilenet | yolov3-lite | 320 | float32 | 57 ms | TensorRT-YOLOv3 | Detection |
mobilenetv2 | yolov3-lite | 320 | float32 | 37 ms | TensorRT-YOLOv3 | Detection |
- pre/post process cosume 10 ms per frame
- Framework link : TensorRT-YOLOv3 , ncnn , darknet, TensorRT-Pelee
- The performance in jetson nano was not satisfactory , I will try to optimize it