Help me, binary segmentation acc error! #2628
Comments
|
would you like to provide the full config about dataset setting? |
# dataset settings
dataset_type = 'ManipulationDataset' #change
data_root = '/home/featurize/data/manipulation'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
crop_size = (512, 512)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='LoadAnnotations'),
dict(type='Resize', img_scale=(1280, 640), ratio_range=(0.5, 2.0)),
dict(type='RandomCrop', crop_size=crop_size, cat_max_ratio=0.75),
dict(type='RandomFlip', prob=0.5),
dict(type='PhotoMetricDistortion'),
dict(type='Normalize', **img_norm_cfg),
dict(type='Pad', size=crop_size, pad_val=0, seg_pad_val=255),
dict(type='DefaultFormatBundle'),
dict(type='Collect', keys=['img', 'gt_semantic_seg']),
]
test_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='MultiScaleFlipAug',
img_scale=(2560, 640),
# img_ratios=[0.5, 0.75, 1.0, 1.25, 1.5, 1.75],
flip=False,
transforms=[
dict(type='Resize', keep_ratio=True),
dict(type='RandomFlip'),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='Collect', keys=['img']),
])
]
data = dict(
samples_per_gpu=6,
workers_per_gpu=4,
train=dict(
type=dataset_type,
data_root=data_root,
img_dir='images/training',
ann_dir='annotations/training',
pipeline=train_pipeline),
val=dict(
type=dataset_type,
data_root=data_root,
img_dir='images/validation',
ann_dir='annotations/validation',
pipeline=test_pipeline),
test=dict(
type=dataset_type,
data_root=data_root,
img_dir='images/validation',
ann_dir='annotations/validation',
pipeline=test_pipeline))
optimizer = dict(type='SGD', lr=0.005, momentum=0.9, weight_decay=0.0005) |
|
Hi @xiaoaxiaoxiaocao, |
|
@xiexinch, the full config and some example images : |
|
Did you use the annotations in data_process for training? It seems that all the annotation images are pure black. |
|
Yes, I used the data of data_process for training. I refer to the drive dataset, and the annotation imgs(data_ori) value divided by 128 is equivalent to '1 if value >= 128 else 0' |
|
what do you mean by "divided by 128 is equivalent to '1 if value >= 128 else 0'"? Do you mind providing the complete preprocessing code used |
|
preprocessing code: for i, file in enumerate(files): |
|
@xiexinch @csatsurnh If I do not do the above data preprocessing, the following error will be reported: 2023-03-05 10:38:24,533 - mmseg - INFO - workflow: [('train', 1)], max: 80000 iters import torch ConvolutionParams terminate called after throwing an instance of 'c10::CUDAError' 已放弃 (核心已转储) |
Hi @xiaoaxiaoxiaocao, |
|
Why? I refer to the preprocessing of the drive dataset, and its preprocessing was done in this way |
We don't know whether your data is the same as drive, so there is no guarantee that drive's processing will work on your dataset as well. |
|
Closing the issue, as there is no activity for a while. |
The custom data set (modeled after the drive data set), the categories are foreground and background. Tannotation imgs value divided by 128 is equivalent to '1 if value >= 128 else 0', the training results are as follows， I don't know how to improve，help me, thanks!
My config:
The text was updated successfully, but these errors were encountered: