Closed KPeng9510 closed 4 years ago
You could simply remove the car related config for a pure detection of pedestrian and/or cyclist. The results of pedestrian/cyclist of KITTI dataset are not stable and need to be further tuned for a better performance. The configs we provided focus on a higher performance of car, we simply add the predictions of three classes for a full detection results, and do not tune the ped/cyc KITTI results with this new repo yet.
Is there any pretrained file which considers also the other classes of KITTI, for example Tram and Truck ?
Hi, could you provide a template of the config file just for pedestrain and cyclist (pointpillars)? The test result for pedestrian which is trained together with car and cyclist is a little worse. Thank you very much!