Is there a conflict between class_weight and seg_ pad_val？
Author: SquirrelEdisonCreated Dec 25, 2024Updated Dec 18, 2025
I fine tuned the segnext model on a custom dataset (including 6 types of backgrounds) and found that the value of the label was normal in the loader (from 0 to num_classes-1), but when passed into the class_weight calculation, there was a sudden addition of 255 values in the label. After debugging, I found that these values were caused by seg_pad_val.
label's val is error-free in the dataloader
label's val is added some '255' val
and in class_weight calculation (/conda/lib/python3.9/site-packages/mmsegmentation-1.2.2-py3.9.egg/mmseg/models/losses/cross_entropy_loss.py):
if (avg_factor is None) and reduction == 'mean':
if class_weight is None:
for cls in label:
print("cls: ", cls)
if avg_non_ignore:
avg_factor = label.numel() - (label
== ignore_index).sum().item()
else:
avg_factor = label.numel()
else:
# the average factor should take the class weights into account
# print("label: ", label)
# print("label_shape: ", label.shape)
# print("class_weight: ", class_weight)
# print("cls in label:")
# for cls in label:
# print("cls: ", cls)
# print("class_weight[cls]: ",class_weight[cls])
label_weights = torch.stack([class_weight[cls] for cls in label
]).to(device=class_weight.device)
if avg_non_ignore:
label_weights[label == ignore_index] = 0
avg_factor = label_weights.sum()label_weights = torch.stack([class_weight[cls] for cls in label ]).to(device=class_weight.device)
the cls in class_weight[cls] will be '255', and this code will cause index out of bound
when I changed the val of seg_pad_val from '255' to '0', the label's val was back to normal
segnext model config file:
_base_ = [
'../_base_/default_runtime.py', '../_base_/schedules/schedule_20k.py',
'../_base_/datasets/fire6dataset.py'
]
# model settings
checkpoint_file = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segnext/mscan_t_20230227-119e8c9f.pth' # noqa
ham_norm_cfg = dict(type='GN', num_groups=32, requires_grad=True)
crop_size = (512, 512)
data_preprocessor = dict(
type='SegDataPreProcessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True,
pad_val=0,
seg_pad_val=255,
size=(512, 512),
test_cfg=dict(size_divisor=32))
model = dict(
type='EncoderDecoder',
data_preprocessor=data_preprocessor,
pretrained=None,
backbone=dict(
type='MSCAN',
init_cfg=dict(type='Pretrained', checkpoint=checkpoint_file),
embed_dims=[32, 64, 160, 256],
mlp_ratios=[8, 8, 4, 4],
drop_rate=0.0,
drop_path_rate=0.1,
depths=[3, 3, 5, 2],
attention_kernel_sizes=[5, [1, 7], [1, 11], [1, 21]],
attention_kernel_paddings=[2, [0, 3], [0, 5], [0, 10]],
act_cfg=dict(type='GELU'),
norm_cfg=dict(type='BN', requires_grad=True)),
decode_head=dict(
type='LightHamHead',
in_channels=[64, 160, 256],
in_index=[1, 2, 3],
channels=256,
ham_channels=256,
dropout_ratio=0.1,
num_classes=6, # include background
norm_cfg=ham_norm_cfg,
align_corners=False,
loss_decode=[
dict(type='CrossEntropyLoss', loss_name='loss_ce', use_sigmoid=False, loss_weight=1.0, class_weight=[0.87, 1.0, 1.85, 1.85, 1.17, 2.14]),
# dict(type='CrossEntropyLoss', loss_name='loss_ce', use_sigmoid=False, loss_weight=1.0),
dict(type='DiceLoss', loss_name='loss_dice', loss_weight=1.0)
],
ham_kwargs=dict(
MD_S=1,
MD_R=16,
train_steps=6,
eval_steps=7,
inv_t=100,
rand_init=True)),
# model training and testing settings
train_cfg=dict(),
test_cfg=dict(mode='whole'))
# dataset settings
train_dataloader = dict(batch_size=8)
# optimizer
optim_wrapper = dict(
_delete_=True,
type='OptimWrapper',
optimizer=dict(
type='AdamW', lr=0.00006, betas=(0.9, 0.999), weight_decay=0.01),
paramwise_cfg=dict(
custom_keys={
'pos_block': dict(decay_mult=0.),
'norm': dict(decay_mult=0.),
'head': dict(lr_mult=10.)
}))
param_scheduler = [
dict(
type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1500),
dict(
type='PolyLR',
power=1.0,
begin=1500,
end=20000,
eta_min=0.0,
by_epoch=False,
)
]_base_/datasets/fire6dataset.py:
# dataset settings
# reduce_zero_label = False
dataset_type = 'Fire6Dataset'
data_root = 'data/ade/FireData'
crop_size = (512, 512)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='LoadAnnotations', reduce_zero_label=False),
dict(
type='RandomResize',
scale=(2048, 512),
ratio_range=(0.5, 2.0),
keep_ratio=True),
dict(type='RandomCrop', crop_size=crop_size, cat_max_ratio=0.75),
dict(type='RandomFlip', prob=0.5),
# dict(type='PhotoMetricDistortion'),
dict(type='PackSegInputs')
]
test_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='Resize', scale=(2048, 512), keep_ratio=True),
# add loading annotation after ``Resize`` because ground truth
# does not need to do resize data transform
dict(type='LoadAnnotations', reduce_zero_label=False),
dict(type='PackSegInputs')
]
img_ratios = [0.5, 0.75, 1.0, 1.25, 1.5, 1.75]
tta_pipeline = [
dict(type='LoadImageFromFile', backend_args=None),
dict(
type='TestTimeAug',
transforms=[
[
dict(type='Resize', scale_factor=r, keep_ratio=True)
for r in img_ratios
],
[
dict(type='RandomFlip', prob=0., direction='horizontal'),
dict(type='RandomFlip', prob=1., direction='horizontal')
], [dict(type='LoadAnnotations')], [dict(type='PackSegInputs')]
])
]
train_dataloader = dict(
batch_size=4,
num_workers=4,
persistent_workers=True,
sampler=dict(type='InfiniteSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_prefix=dict(
img_path='images/training', seg_map_path='annotations/training'),
pipeline=train_pipeline))
val_dataloader = dict(
batch_size=1,
num_workers=4,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_prefix=dict(
img_path='images/validation',
seg_map_path='annotations/validation'),
pipeline=test_pipeline))
test_dataloader = val_dataloader
val_evaluator = dict(type='IoUMetric', iou_metrics=['mIoU'])
test_evaluator = val_evaluatormmseg/datasets/fire_dataset.py:
# Copyright (c) OpenMMLab. All rights reserved.
from mmseg.registry import DATASETS
from .basesegdataset import BaseSegDataset
@DATASETS.register_module()
class FireDataset(BaseSegDataset):
"""ADE20K dataset.
In segmentation map annotation for ADE20K, 0 stands for background, which
is not included in 150 categories. ``reduce_zero_label`` is fixed to True.
The ``img_suffix`` is fixed to '.jpg' and ``seg_map_suffix`` is fixed to
'.png'.
"""
METAINFO = dict(
classes=('fire','smoke_black','smoke_white','smoke_yellow','spark'),
palette=[[120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50],
[4, 200, 3]])
def __init__(self,
img_suffix='.jpg',
seg_map_suffix='.png',
reduce_zero_label=True,
**kwargs) -> None:
super().__init__(
img_suffix=img_suffix,
seg_map_suffix=seg_map_suffix,
reduce_zero_label=reduce_zero_label,
**kwargs)Source: open-mmlab/mmsegmentation