23 lines (19 loc) · 549 Bytes

Convert pytorch to tensorrt (fp16)

How to Run

train base model with imagenet100 dataset
- Train Base Model (resnet18)
export onnx model
```
python onnx_export.py
```
generate tensorrt model
```
python onnx2trt.py
```

fp16
Gpu Mem: 138M
[TRT_E] Test Top-1 Accuracy: 84.54 [%]
[TRT_E] Test Top-5 Accuracy: 97.20 [%]
[TRT_E] 10000 iterations time: 6.83 [sec]
[TRT_E] Average FPS: 1463.45 [fps]
[TRT_E] Average inference time: 0.68 [msec]