Pretrained model for Ground-aware Monocular 3D Object Detection for Autonomous Driving.
the model file could be placed under workdirs/Stereo3D/checkpoint/ (you should provide the path to the model file in the command line)
anchor_mean/std_Car/Pedestrian.npy should be placed under workdirs/Yolo3D/output/training. You can reproduce the npy file with the scripts runned on the 'test split'.
Backward Incompatibility:
We update the function for converting between observation angle (alpha) and 3D rotation angle (theta) following the more accurate version from RTM3D. It will break the result of the models in the previous release.
And we have to retrain a new YOLOStereo3D model to adapt to this change. So the released model performs slightly differently from the KITTI one.
Notice:
To get similar performance on the test-split, you need to train for more epochs (80 epochs test for example), while you only need about 50 epochs to get a saturated performance on validation split (empirically with the current learning rate settings).
| Benchmark | Easy | Moderate | Hard |
|---|---|---|---|
| Car Detection | 94.75 % | 84.50 % | 62.13 % |
| Car Orientation | 93.65 % | 82.88 % | 60.92 % |
| Car 3D Detection | 65.77 % | 40.71 % | 29.99 % |
| Car Bird's Eye View | 74.00 % | 49.54 % | 36.30 % |
| Pedestrian Detection | 58.34 % | 49.54 % | 36.30 % |
| Pedestrian Orientation | 50.41 % | 36.81 % | 31.51 % |
| Pedestrian 3D Detection | 31.03 % | 20.67 % | 18.34 % |
| Pedestrian Bird's Eye View | 32.52 % | 22.74 % | 19.16 % |