PYTHONWARNINGS="ignore" CUDA_VISIBLE_DEVICES=4 python -m torch.distributed.launch --nproc_per_node=1 \
    --master_port=$(( RANDOM % 1000 + 50000 )) \
    eval_sfrustum.py --config configs/sfrustum_scannet.txt \
    --ckpt out/frustum-scannet-generalized_trainning_4layer/model_379999.pth \
    --expname frustum-scannet-eval --val_set_list configs/scannetv2_test_split.txt --no_load_opt --no_load_scheduler