search

NVIDIA-AI-IOT / Lidar_AI_Solution

A project demonstrating Lidar related AI solutions, including three GPU accelerated Lidar/camera DL networks (PointPillars, CenterPoint, BEVFusion) and the related libs (cuPCL, 3D SparseConvolution, YUV2RGB, cuOSD,).
Other
1.33k stars 232 forks source link

fps result #223

Open Yunge6666 opened 11 months ago

Yunge6666 commented 11 months ago

[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 81/81, 5.4 task/s, elapsed: 15s, ETA: 0sStep1 Done. Start to convert detection format... [>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 81/81, 86.0 task/s, elapsed: 1s, ETA: 0s Results writes to /tmp/tmpwb0m71nk/results/results_nusc.json mAP: 0.5716
mATE: 0.4096 mASE: 0.4423 mAOE: 0.4780 mAVE: 0.4115 mAAE: 0.3107 NDS: 0.5806 Eval time: 1.2s

Per-class results: Object Class AP ATE ASE AOE AVE AAE
car 0.925 0.170 0.156 0.101 0.119 0.070 truck 0.804 0.157 0.127 0.068 0.071 0.000 bus 0.994 0.173 0.076 0.021 0.556 0.302 trailer 0.000 1.000 1.000 1.000 1.000 1.000 construction_vehicle 0.000 1.000 1.000 1.000 1.000 1.000 pedestrian 0.933 0.125 0.254 0.331 0.215 0.114 motorcycle 0.736 0.191 0.257 0.400 0.053 0.000 bicycle 0.589 0.198 0.227 0.381 0.278 0.000 traffic_cone 0.734 0.082 0.325 nan nan nan
barrier 0.000 1.000 1.000 1.000 nan nan

This is the result I got, but I don't understand how fps is calculated? What do the task/s and eval time here represent respectively? Could anyone please explain it, I would be very grateful.

hopef commented 11 months ago

fps is computed by CUDA-BEVFusion on ORIN.

Yunge6666 commented 11 months ago

fps is computed by CUDA-BEVFusion on ORIN.

Do you provide any code to calculate FPS in the CUDA-BEVfusion? In addition, do you understand what task/s and eval time are in my results above?

hopef commented 11 months ago

https://github.com/NVIDIA-AI-IOT/Lidar_AI_Solution/blob/a8461f0a024477dbcf8746a3ea8a0f5e3aa14540/CUDA-BEVFusion/src/main.cpp#L258C6-L258C6