Skip to content

Latest commit

 

History

History
55 lines (44 loc) · 3.03 KB

MODEL.md

File metadata and controls

55 lines (44 loc) · 3.03 KB
Raw

Training

cd tools
sh scripts/dist_train.sh ${NUM_GPUS} --cfg_file ${CONFIG_FILE} ${PY_ARGS}

For instance,

  • train B15/N4 semantic segmentation on ScanNet: 
    cd tools
sh scripts/dist_train.sh 8 --cfg_file cfgs/scannet_models/spconv_clip_base15_caption_adamw.yaml --extra_tag exp_tag
  • train B13/N4 instance segmentation on ScanNet: 
    cd tools
sh scripts/dist_train.sh 8 --cfg_file cfgs/scannet_models/inst/softgroup_clip_base13_caption_adamw.yaml --extra_tag exp_tag

Inference

cd tools
sh scripts/dist_test.sh ${NUM_GPUS} --cfg_file ${CONFIG_FILE} --ckpt ${CKPT_PATH}

For instance,

  • to test a B15/N4 model on ScanNet: 
    cd tools
sh scripts/dist_test.sh 8 --cfg_file cfgs/scannet_models/spconv_clip_base15_caption_adamw.yaml --ckpt output/scannet_models/spconv_clip_base15_caption/exp_tag/ckpt/checkpoint_ep128.pth

Model Zoo

  • semantic segmentation

    Dataset Partition hIoU / mIoU(B) / mIoU(N) Path
    ScanNet B15/N4 64.9 / 67.8 / 62.2 ckpt
    ScanNet B12/N7 55.9 / 70.4 / 46.4 ckpt
    ScanNet B10/N9 52.8 / 76.6 / 40.3 ckpt
    S3DIS B8/N4 35.6 / 58.3 / 25.6 ckpt
    S3DIS B6/N6 38.4 / 53.9 / 29.8 ckpt

  • instance segmentation

    Dataset Partition hAP50 / mAP50(B) / mAP50(N) Path
    ScanNet B13/N4 57.8 / 58.7 / 56.9 ckpt
    ScanNet B10/N7 31.6 / 54.8 / 22.2 ckpt
    ScanNet B8/N9 36.9 / 63.1 / 26.2 ckpt
    S3DIS B8/N4 17.2 / 60.9 / 10.0 ckpt
    S3DIS B6/N6 15.8 / 48.2 / 9.5 ckpt