PTv1 config for outdoor datasets

[gitKincses]

I'm testing the outdoor dataset DALES in PTv1. However, I am encountering some issues. There are significant fluctuations at the beginning of the test, as shown in the chart.

I'm not sure if this is due to using an unsuitable config, which I directly moved from s3dis/semseg-pt-v1-0-base.py.

_base_ = ["../_base_/default_runtime.py"]
# misc custom setting
# batch_size = 12  # bs: total bs in all gpus
batch_size = 12  # bs: total bs in all gpus
mix_prob = 0.8
empty_cache = False
enable_amp = True

# model settings
model = dict(
    type="DefaultSegmentor",
    backbone=dict(
        type="PointTransformer-Seg26",
        in_channels=4,
        # num_classes=8,
        num_classes=8,
    ),
    criteria=[dict(type="CrossEntropyLoss", loss_weight=1.0, ignore_index=-1)],
)

# scheduler settings
epoch = 100
optimizer = dict(type="AdamW", lr=0.006, weight_decay=0.05)
scheduler = dict(type="MultiStepLR", milestones=[0.6, 0.8], gamma=0.1)

# dataset settings
dataset_type = "DALESDataset"
data_root = "/home/Pointcept/data/dales"

data = dict(
    # num_classes=8,
    num_classes=8,
    ignore_index=-1,
    names=[
        "Ground",
        "Vegetation",
        "Cars",
        "Trucks",
        "Power lines",
        "Fences",
        "Poles",
        "Buildings"
    ],
    train=dict(
        type=dataset_type,
        split="train",
        data_root=data_root,
        transform=[
            # dict(type="RandomDropout", dropout_ratio=0.2, dropout_application_ratio=0.2),
            # dict(type="RandomRotateTargetAngle", angle=(1/2, 1, 3/2), center=[0, 0, 0], axis="z", p=0.75),
            dict(type="RandomRotate", angle=[-1, 1], axis="z", center=[0, 0, 0], p=0.5),
            # dict(type="RandomRotate", angle=[-1/6, 1/6], axis="x", p=0.5),
            # dict(type="RandomRotate", angle=[-1/6, 1/6], axis="y", p=0.5),
            dict(type="PointClip", point_cloud_range=(-75.2, -75.2, -4, 75.2, 75.2, 2)),
            dict(type="RandomScale", scale=[0.9, 1.1]),
            # dict(type="RandomShift", shift=[0.2, 0.2, 0.2]),
            dict(type="RandomFlip", p=0.5),
            dict(type="RandomJitter", sigma=0.005, clip=0.02),
            # dict(type="ElasticDistortion", distortion_params=[[0.2, 0.4], [0.8, 1.6]]),
            dict(
                type="GridSample",
                grid_size=0.05,
                hash_type="fnv",
                mode="train",
                keys=("coord", "strength", "segment"),
                return_grid_coord=True,
            ),
            dict(type="SphereCrop", point_max=80000, mode="random"),
            # dict(type="CenterShift", apply_z=False),
            dict(type="ToTensor"),
            dict(
                type="Collect",
                keys=("coord", "grid_coord", "segment"),
                feat_keys=("coord", "strength"),
            ),
        ],
        test_mode=False,
        ignore_index=-1,
    ),
    val=dict(
        type=dataset_type,
        split="test",
        data_root=data_root,
        transform=[
            dict(type="PointClip", point_cloud_range=(-75.2, -75.2, -4, 75.2, 75.2, 2)),
            dict(
                type="GridSample",
                grid_size=0.05,
                hash_type="fnv",
                mode="train",
                keys=("coord", "strength", "segment"),
                return_grid_coord=True,
            ),
            dict(type="ToTensor"),
            dict(
                type="Collect",
                keys=("coord", "grid_coord", "segment"),
                feat_keys=("coord", "strength"),
            ),
        ],
        test_mode=False,
        ignore_index=-1,
    ),
    test=dict(
        type=dataset_type,
        split="test",
        data_root=data_root,
        transform=[
            dict(type="PointClip", point_cloud_range=(-75.2, -75.2, -4, 75.2, 75.2, 2)),
            dict(type="Copy", keys_dict={"segment": "origin_segment"}),
            dict(
                type="GridSample",
                grid_size=0.025,
                hash_type="fnv",
                mode="train",
                keys=("coord", "strength", "segment"),
                return_inverse=True,
            ),
        ],
        test_mode=True,
        test_cfg=dict(
            voxelize=dict(
                type="GridSample",
                grid_size=0.05,
                hash_type="fnv",
                mode="test",
                return_grid_coord=True,
                keys=("coord", "strength"),
            ),
            crop=None,
            post_transform=[
                dict(type="ToTensor"),
                dict(
                    type="Collect",
                    keys=("coord", "grid_coord", "index"),
                    feat_keys=("coord", "strength"),
                ),
            ],
            aug_transform=[
                [
                    dict(
                        type="RandomRotateTargetAngle",
                        angle=[0],
                        axis="z",
                        center=[0, 0, 0],
                        p=1,
                    )
                ],
                [
                    dict(
                        type="RandomRotateTargetAngle",
                        angle=[1 / 2],
                        axis="z",
                        center=[0, 0, 0],
                        p=1,
                    )
                ],
                [
                    dict(
                        type="RandomRotateTargetAngle",
                        angle=[1],
                        axis="z",
                        center=[0, 0, 0],
                        p=1,
                    )
                ],
                [
                    dict(
                        type="RandomRotateTargetAngle",
                        angle=[3 / 2],
                        axis="z",
                        center=[0, 0, 0],
                        p=1,
                    )
                ],
            ],
        ),
        ignore_index=-1,
    ),
)

[Gofinge]

Hi, please use PTv3 and PTv2 instead. We didn't optimized PTv1 for outdoor lidar dataset.