Pretrained image backbone checkpoints can be downloaded from HERE. Move img_backbone.pth and img_backbone_sem.pth under the directory mmdet3d/models/backbones/.
Online 3D semantic segmentation
Train and evaluate MinkUNet on ScanNet-SV:
CUDA_VISIBLE_DEVICES=0,1 PORT=29544 bash ./tools/dist_train.sh configs/minkunet_sv/minkunetFF_scannet_seg-3d-20class.py 2 --work-dir work_dirs/minkunet_svFF
CUDA_VISIBLE_DEVICES=0 PORT=29544 bash ./tools/dist_test.sh configs/minkunet_sv/minkunetFF_scannet_seg-3d-20class.py work_dirs/minkunet_svFF/latest.pth 1 --show-dir work_dirs/debug/ --eval x
Finetune and evaluate MinkUNet (with adapters) on ScanNet-MV:
CUDA_VISIBLE_DEVICES=0,1 PORT=29544 bash ./tools/dist_train.sh configs/minkunet_online/minkunetFF_online_scannet_seg-3d-20class.py 2 --work-dir work_dirs/minkunet_online
CUDA_VISIBLE_DEVICES=0 PORT=29544 bash ./tools/dist_test.sh configs/minkunet_online/minkunetFF_online_scannet_seg-3d-20class.py work_dirs/minkunet_online/latest.pth 1 --show-dir work_dirs/debug/ --eval x
Evaluate MinkUNet (with adapters) on SceneNN-MV:
CUDA_VISIBLE_DEVICES=0 PORT=29544 bash ./tools/dist_test.sh configs/minkunet_online/minkunetFF_scenenn_seg-3d-16class.py work_dirs/minkunet_online/latest.pth 1 --show-dir work_dirs/debug/ --eval x
Visualize MinkUNet (with adapters) on ScanNet-MV:
CUDA_VISIBLE_DEVICES=0 PORT=29544 bash ./tools/dist_test.sh configs/minkunet_online/minkunetFF_online_scannet_seg-3d-20class.py work_dirs/minkunet_online/latest.pth 1 --show-dir work_dirs/vis --eval x --show
Online 3D object detection
Train and evaluate FCAF3D on ScanNet-SV:
CUDA_VISIBLE_DEVICES=0,1 PORT=29544 bash ./tools/dist_train.sh configs/fcaf3d_sv/fcaf3dFF_scannet-3d-18class.py 2 --work-dir work_dirs/fcaf3d_svFF
CUDA_VISIBLE_DEVICES=0 PORT=29544 bash ./tools/dist_test.sh configs/fcaf3d_sv/fcaf3dFF_scannet-3d-18class.py work_dirs/fcaf3d_svFF/latest.pth 1 --show-dir work_dirs/debug/ --eval x
Finetune and evaluate FCAF3D (with adapters) on ScanNet-MV:
CUDA_VISIBLE_DEVICES=0,1 PORT=29544 bash ./tools/dist_train.sh configs/fcaf3d_online/online3dv3FF_8x2_scannet-3d-18class.py 2 --work-dir work_dirs/online3d_v3FF
CUDA_VISIBLE_DEVICES=0 PORT=29544 bash ./tools/dist_test.sh configs/fcaf3d_online/online3dv3FF_8x2_scannet-3d-18class.py work_dirs/online3d_v3FF/latest.pth 1 --show-dir work_dirs/debug/ --eval x
Visualize FCAF3D (with adapters) on ScanNet-MV:
CUDA_VISIBLE_DEVICES=0 PORT=29544 bash ./tools/dist_test.sh configs/fcaf3d_online/online3dv3FF_8x2_scannet-3d-18class.py work_dirs/online3d_v3FF/latest.pth 1 --show-dir work_dirs/vis --eval x --show
Online 3D instance segmentation
Train and evaluate TD3D on ScanNet-SV:
CUDA_VISIBLE_DEVICES=0,1 PORT=29544 bash ./tools/dist_train.sh configs/td3d_sv/td3dFF_is_scannet-3d-18class.py 2 --work-dir work_dirs/td3d_svFF
CUDA_VISIBLE_DEVICES=0 PORT=29544 bash ./tools/dist_test.sh configs/td3d_sv/td3dFF_is_scannet-3d-18class.py work_dirs/td3d_svFF/latest.pth 1 --show-dir work_dirs/debug/ --eval x
Finetune and evaluate TD3D (with adapters) on ScanNet-MV:
CUDA_VISIBLE_DEVICES=0,1 PORT=29544 bash ./tools/dist_train.sh configs/td3d_online/td3dFF_online_is_scannet-3d-18class.py 2 --work-dir work_dirs/td3d_online
CUDA_VISIBLE_DEVICES=0 PORT=29544 bash ./tools/dist_test.sh configs/td3d_online/td3dFF_online_is_scannet-3d-18class.py work_dirs/td3d_online/latest.pth 1 --show-dir work_dirs/debug/ --eval x
Visualize TD3D (with adapters) on ScanNet-MV:
CUDA_VISIBLE_DEVICES=0 PORT=29544 bash ./tools/dist_test.sh configs/td3d_online/td3dFF_online_is_scannet-3d-18class.py work_dirs/td3d_online/latest.pth 1 --show-dir work_dirs/vis --eval x --show