Our code will be released soon.

Framework Overview

BEV Features Generation

Use TAKD

Installation

Please follow INSTALL to install CMKD.

Getting Started

Please follow GETTING_START to train or evaluate the models.

Models

KITTI

Waymo

Coming Soon

Nuscenes

=======

MonoTAKD

Paper : MonoTAKD: Teaching assistant knowledge distillation for monocular 3D object detection.

Train python train_TAKD.py --cfg cfgs/kitti_models/TAKD/TAKD-scd/kitti_R50_scd_TAKD.yaml --pretrained_lidar_model ../checkpoints/scd-teacher-kitti.pth --pretrained_img_model ../checkpoints/cmkd-scd-2161.pth

Test python test_TAKD.py --cfg cfgs/kitti_models/TAKD/TAKD-scd/kitti_R50_scd_TAKD.yaml --ckpt ../checkpoints/twcc_8015_checkpoint_epoch_10.pth

Train Teacher: python train.py --cfg cfgs/kitti_models/second_teacher.yaml --ckpt ../checkpoints/scd-teacher-kitti.pth --pretrained_model ../checkpoints/scd-teacher-kitti.pth python train.py --cfg cfgs/kitti_models/second.yaml (w/o pretrained)

Test Teacher: python test.py --cfg cfgs/kitti_models/second_teacher.yaml --ckpt ../checkpoints/scd-teacher-kitti.pth --save_to_file

Tensorboard: tensorboard --logdir ../output/kitti_models/TAKD/TAKD-scd/kitti_R50_scd_TAKD/

9d852a3f193097eb414f776304cb2d8b7fb5e1de