Skip to content

Latest commit

 

History

History
59 lines (38 loc) · 3.5 KB

README.md

File metadata and controls

59 lines (38 loc) · 3.5 KB
Raw

MODEL ZOO

Common settings and notes

  • The experiments are run with PyTorch 1.1, CUDA 10.0, and CUDNN 7.5.
  • The training is conducted on 4 V100 GPUs in a DGX server.
  • Testing times are measured on a TITAN RTX GPU with batch size 1.

nuScenes 3D Detection

We provide training / validation configurations, logs, pretrained models, and prediction files for all models in the paper

VoxelNet

Model Validation MAP Validation NDS Link
centerpoint_voxel_1440 59.6 66.8 URL

VoxelNet(depreacted)

These results are obtained before the bug fix.

Model FPS Validation MAP Validation NDS Link
centerpoint_voxel_1024 16 56.4 64.8 URL
centerpoint_voxel_1440_dcn 11 57.1 65.4 URL
centerpoint_voxel_1440_dcn(flip) 3 59.5 67.4 URL

PointPillars

Model FPS Validation MAP Validation NDS Link
centerpoint_pillar 31 50.3 60.2 URL

nuScenes 3D Tracking

Model Tracking time Total time Validation AMOTA ↑ Validation AMOTP ↓ Link
centerpoint_voxel_1024 1ms 64ms 63.7 0.606 URL
centerpoint_voxel_1440_dcn 1ms 95ms 64.1 0.596 URL
centerpoint_voxel_1440_dcn(flip test) 1ms 343ms 66.5 0.567 URL

nuScenes test set Detection/Tracking

Detection

Model Test MAP Test NDS Link
centerpoint_voxel_1440_dcn 58.0 65.5 Detection

Tracking

Model Test AMOTA Test AMOTP Link
centerpoint_voxel_1440_dcn(flip test) 63.8* 0.555* Detection / Tracking

*The numbers are from an old version of the codebase. Current detection models should perform slightly better.