| # a indoor_ds model with the pos_enc impl bug fixed. | |
| SCRIPTPATH=$(dirname $(readlink -f "$0")) | |
| PROJECT_DIR="${SCRIPTPATH}/../../" | |
| # conda activate loftr | |
| export PYTHONPATH=$PROJECT_DIR:$PYTHONPATH | |
| cd $PROJECT_DIR | |
| data_cfg_path="configs/data/scannet_test_1500.py" | |
| main_cfg_path="configs/aspan/indoor/aspan_test.py" | |
| ckpt_path='weights/indoor.ckpt' | |
| dump_dir="dump/indoor_dump" | |
| profiler_name="inference" | |
| n_nodes=1 # mannually keep this the same with --nodes | |
| n_gpus_per_node=-1 | |
| torch_num_workers=4 | |
| batch_size=1 # per gpu | |
| python -u ./test.py \ | |
| ${data_cfg_path} \ | |
| ${main_cfg_path} \ | |
| --ckpt_path=${ckpt_path} \ | |
| --dump_dir=${dump_dir} \ | |
| --gpus=${n_gpus_per_node} --num_nodes=${n_nodes} --accelerator="ddp" \ | |
| --batch_size=${batch_size} --num_workers=${torch_num_workers}\ | |
| --profiler_name=${profiler_name} \ | |
| --benchmark \ | |
| --mode integrated | |