Skip to content

run.md

Latest commit

 

History

History
69 lines (46 loc) · 1.93 KB

run.md

File metadata and controls

69 lines (46 loc) · 1.93 KB
Raw

Run and Eval

Training

All configs are located under projects/configs/sgn/.

SemanticKITTI (with MobileStereoNet depth)

./tools/dist_train.sh projects/configs/sgn/proood-semkitti.py 4

SemanticKITTI (with SQL depth)

./tools/dist_train.sh projects/configs/sgn/proood-sql-semkitti.py 4

KITTI-360 (with MobileStereoNet depth)

./tools/dist_train.sh projects/configs/sgn/proood-kitti360.py 4

KITTI-360 (with SQL depth)

./tools/dist_train.sh projects/configs/sgn/proood-sql-kitti360.py 4

Evaluation

Replace ./path/to/ckpts.pth with your actual checkpoint path.

SemanticKITTI

./tools/dist_test.sh projects/configs/sgn/proood-semkitti.py ./path/to/ckpts.pth 4
./tools/dist_test.sh projects/configs/sgn/proood-sql-semkitti.py ./path/to/ckpts.pth 4

KITTI-360

./tools/dist_test.sh projects/configs/sgn/proood-kitti360.py ./path/to/ckpts.pth 4
./tools/dist_test.sh projects/configs/sgn/proood-sql-kitti360.py ./path/to/ckpts.pth 4

OOD Evaluation

After training, evaluate OOD detection capability using OccOoD benchmarks:

# Vaakitti (OOD objects in SemanticKITTI)
./tools/dist_test_ood.sh projects/configs/sgn/proood-ood-vaakitti.py ./path/to/ckpts.pth 4

# Vaakitti360 (OOD objects in KITTI-360)
./tools/dist_test_ood.sh projects/configs/sgn/proood-ood-vaakitti360.py ./path/to/ckpts.pth 4

# STU (cross-dataset OOD)
./tools/dist_test_ood.sh projects/configs/sgn/proood-ood-stu.py ./path/to/ckpts.pth 4

Notes

  • The 4 at the end of each command specifies the number of GPUs to use. Adjust based on available hardware.
  • Config paths use forward slashes and are relative to the project root.
  • Before training, update data_root in the config file to point to your local dataset location.
  • ResNet-50 pretrained weights are expected at ckpts/resnet50-19c8e357.pth.