PTV3 config for SemanticKitti

[Charlie839242]

Hi, thanks for your great work!
Recently I am trying to train PTv3 on SemanticKitti dataset. However, the train loss is always 0 while the miou keeps around 5% since the first epoch. Could you please help to see whether there is something wrong with the config? Thanks!
Looking forward to your reply!

Here is the config

weight = None
resume = False
evaluate = True
test_only = False
seed = 114514999
save_path = 'exp/'
num_worker = 24
batch_size = 8
batch_size_val = None
batch_size_test = None
epoch = 50
eval_epoch = 50
sync_bn = True
enable_amp = True
empty_cache = False
find_unused_parameters = False
mix_prob = 0.8
hooks = [
dict(type='CheckpointLoader'),
dict(type='IterationTimer', warmup_iter=2),
dict(type='InformationWriter'),
dict(type='SemSegEvaluator'),
dict(type='CheckpointSaver', save_freq=None),
dict(type='PreciseEvaluator', test_last=False)
]
train = dict(type='DefaultTrainer')
test = dict(type='SemSegTester', verbose=True)

model settings

model = dict(
type="DefaultSegmentorV2",
num_classes=19,
backbone_out_channels=64,
backbone=dict(
type="PT-v3m1",
in_channels=4,
order=["z", "z-trans", "hilbert", "hilbert-trans"],
stride=(2, 2, 2, 2),
enc_depths=(2, 2, 2, 6, 2),
enc_channels=(32, 64, 128, 256, 512),
enc_num_head=(2, 4, 8, 16, 32),
enc_patch_size=(1024, 1024, 1024, 1024, 1024),
dec_depths=(2, 2, 2, 2),
dec_channels=(64, 64, 128, 256),
dec_num_head=(4, 4, 8, 16),
dec_patch_size=(1024, 1024, 1024, 1024),
mlp_ratio=4,
qkv_bias=True,
qk_scale=None,
attn_drop=0.0,
proj_drop=0.0,
drop_path=0.3,
shuffle_orders=True,
pre_norm=True,
enable_rpe=False,
enable_flash=True,
upcast_attention=False,
upcast_softmax=False,
cls_mode=False,
pdnorm_bn=False,
pdnorm_ln=False,
pdnorm_decouple=True,
pdnorm_adaptive=False,
pdnorm_affine=True,
pdnorm_conditions=("nuScenes", "SemanticKITTI", "Waymo"),
),
criteria=[
dict(type="CrossEntropyLoss", loss_weight=1.0, ignore_index=-1),
dict(type="LovaszLoss", mode="multiclass", loss_weight=1.0, ignore_index=-1),
],
)

param_dicts = [dict(keyword="block", lr=0.0002)]
optimizer = dict(type="AdamW", lr=0.002, weight_decay=0.005)
scheduler = dict(
type="OneCycleLR",
max_lr=[0.002, 0.0002],
pct_start=0.04,
anneal_strategy="cos",
div_factor=10.0,
final_div_factor=100.0,
)

dataset settings

dataset_type = "SemanticKITTIDataset"
data_root = "data/semantic_kitti"
ignore_index = -1
names = [
"car",
"bicycle",
"motorcycle",
"truck",
"other-vehicle",
"person",
"bicyclist",
"motorcyclist",
"road",
"parking",
"sidewalk",
"other-ground",
"building",
"fence",
"vegetation",
"trunk",
"terrain",
"pole",
"traffic-sign",
]

data = dict(
num_classes=19,
ignore_index=ignore_index,
names=names,
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="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="PointClip", point_cloud_range=(-35.2, -35.2, -4, 35.2, 35.2, 2)),
dict(type="SphereCrop", sample_rate=0.8, mode="random"),
dict(type="SphereCrop", point_max=120000, 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=ignore_index,
),
val=dict(
type=dataset_type,
split="val",
data_root=data_root,
transform=[
dict(
type="GridSample",
grid_size=0.05,
hash_type="fnv",
mode="train",
keys=("coord", "strength", "segment"),
return_grid_coord=True,
),
dict(type="PointClip", point_cloud_range=(-35.2, -35.2, -4, 35.2, 35.2, 2)),
dict(type="ToTensor"),
dict(
type="Collect",
keys=("coord", "grid_coord", "segment"),
feat_keys=("coord", "strength"),
),
],
test_mode=False,
ignore_index=ignore_index,
),
test=dict(
type=dataset_type,
split="val",
data_root=data_root,
transform=[],
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="PointClip",
point_cloud_range=(-35.2, -35.2, -4, 35.2, 35.2, 2),
),
dict(type="ToTensor"),
dict(
type="Collect",
keys=("coord", "grid_coord", "index"),
feat_keys=("coord", "strength"),
),
],
aug_transform=[
[{
'type': 'RandomScale',
'scale': [0.9, 0.9]
}], [{
'type': 'RandomScale',
'scale': [0.95, 0.95]
}], [{
'type': 'RandomScale',
'scale': [1, 1]
}], [{
'type': 'RandomScale',
'scale': [1.05, 1.05]
}], [{
'type': 'RandomScale',
'scale': [1.1, 1.1]
}],
[{
'type': 'RandomScale',
'scale': [0.9, 0.9]
}, {
'type': 'RandomFlip',
'p': 1
}],
[{
'type': 'RandomScale',
'scale': [0.95, 0.95]
}, {
'type': 'RandomFlip',
'p': 1
}],
[{
'type': 'RandomScale',
'scale': [1, 1]
}, {
'type': 'RandomFlip',
'p': 1
}],
[{
'type': 'RandomScale',
'scale': [1.05, 1.05]
}, {
'type': 'RandomFlip',
'p': 1
}],
[{
'type': 'RandomScale',
'scale': [1.1, 1.1]
}, {
'type': 'RandomFlip',
'p': 1
}]
],
),
ignore_index=ignore_index,
),
)

[Charlie839242]

The tensorboard shows like this

[Xiaoyang Wu]

Working on it soon. Forgive me, as too busy. I receive multiple requests to add support to ModelNet and ScanNet++, as well as other research tasks. I will have some time one month later.