Accelerate Pelee with tensorRT Pelee: A Real-Time Object Detection System on Mobile Devices (NeurIPS 2018)
TensorRT-Pelee can run over 200FPS(5ms) on Titan V and 70FPS(11ms) on jetson TX2(FP32)
Performance:
72 FPS, 13.8 ms (FP32)
GPU Time:
Time by layers:
Requierments:
1.TensorRT4 (Jetpack 3.3 on TX2)
2.CUDA 9.0
3.cudnn7
Run:
cmake .
make
./build/bin/pelee
Reference:
https://github.com/Ghustwb/MobileNet-SSD-TensorRT
TODO:
- FP16 Implementation
- Change Custom layers IPlugin to IPluginExt
The bug has been fixed