DRP-AI¶
The models in NetsPresso Trainer can be converted to DRP-AI format by NetsPresso's Launcher module. The converted DRP-AI model's performance can be measured on actual boards using NetsPresso's Benchmarker module. For more detailed information, please refer to NetsPresso.
Note that the latency value only measures the time for the model's computation and does not include the time on data preprocessing or postprocessing.
Renesas RZ V2L¶
FP16¶
| Task | Model | Input shape | Classes | Latency (ms) | GPU Memory (MB) | CPU Memory (MB) | Ramarks |
|---|---|---|---|---|---|---|---|
| Classification | ResNet18 | (224, 224) | 3 | 23.9425 | - | - | onnx_opset=13 |
| Classification | ResNet34 | (224, 224) | 3 | 40.0319 | - | - | onnx_opset=13 |
| Classification | ResNet50 | (224, 224) | 3 | 58.8135 | - | - | onnx_opset=13 |
| Detection | YOLOX-s | (640, 640) | 4 | 196.642 | - | - | onnx_opset=13 |
| Detection | YOLOX-m | (640, 640) | 4 | 412.707 | - | - | onnx_opset=13 |
| Detection | YOLOX-l | (640, 640) | 4 | 687.293 | - | - | onnx_opset=13 |