Comments (3)
@Liwiman 您的问题缺乏细节,我无法回答,我不知道你有几类,也不知道你 forest land 是第几类,或者其对应的label index是多少,因此我无法给出修改config的意见
from vitae-transformer-remote-sensing.
感谢您的回答我已经解决了上诉问题,并将模型投入训练了,但我在训练过程中出现了问题,想恳请您解答。
我训练的数据集是128*128的100000张遥感图片,在
预训练权重的基础上进行训练,设置batch_size为16训练107500个iter后设置batch_size为40训练了22500个iter,但是模型效果很差(val的图片也是训练的图片,其实是在训练集的部分上val):{"mode": "val", "epoch": 7, "iter": 5129, "lr": 5e-05, "aAcc": 0.6659, "mIoU": 0.4843, "mAcc": 0.6469, "mFscore": 0.6483, "mPrecision": 0.6688, "mRecall": 0.6469, "IoU.water": 0.5942, "IoU.transportation": 0.4205, "IoU.architecture": 0.5802, "IoU.farmland": 0.4988, "IoU.grassland": 0.3906, "IoU.forest land": 0.5741, "IoU.bare soil": 0.3818, "IoU.others": 0.4341, "Acc.water": 0.855, "Acc.transportation": 0.5327, "Acc.architecture": 0.6978, "Acc.farmland": 0.6573, "Acc.grassland": 0.6929, "Acc.forest land": 0.6918, "Acc.bare soil": 0.4493, "Acc.others": 0.5983, "Fscore.water": 0.7455, "Fscore.transportation": 0.592, "Fscore.architecture": 0.7343, "Fscore.farmland": 0.6656, "Fscore.grassland": 0.5617, "Fscore.forest land": 0.7294, "Fscore.bare soil": 0.5526, "Fscore.others": 0.6054, "Precision.water": 0.6608, "Precision.transportation": 0.6662, "Precision.architecture": 0.7749, "Precision.farmland": 0.6741, "Precision.grassland": 0.4723, "Precision.forest land": 0.7715, "Precision.bare soil": 0.7175, "Precision.others": 0.6128, "Recall.water": 0.855, "Recall.transportation": 0.5327, "Recall.architecture": 0.6978, "Recall.farmland": 0.6573, "Recall.grassland": 0.6929, "Recall.forest land": 0.6918, "Recall.bare soil": 0.4493, "Recall.others": 0.5983},就算输入的是训练集图片,推断效果还是很差:
而该图片的label可视化应该是这样的:
我模型的config文件如下:
norm_cfg = dict(type='BN', requires_grad=True)
model = dict(
type='EncoderDecoder',
pretrained=
'work_dirs/cd_train/iter_25000.pth',
backbone=dict(
type='ViTAE_Window_NoShift_basic',
RC_tokens_type=['swin', 'swin', 'transformer', 'transformer'],
NC_tokens_type=['swin', 'swin', 'transformer', 'transformer'],
stages=4,
embed_dims=[64, 64, 128, 256],
token_dims=[64, 128, 256, 512],
downsample_ratios=[4, 2, 2, 2],
NC_depth=[2, 2, 8, 2],
NC_heads=[1, 2, 4, 8],
RC_heads=[1, 1, 2, 4],
mlp_ratio=4.0,
NC_group=[1, 32, 64, 128],
RC_group=[1, 16, 32, 64],
img_size=512,
window_size=7,
drop_path_rate=0.3),
decode_head=dict(
type='UPerHead',
in_channels=[64, 128, 256, 512],
in_index=[0, 1, 2, 3],
pool_scales=(1, 2, 3, 6),
channels=512,
dropout_ratio=0.1,
num_classes=8,
norm_cfg=dict(type='BN', requires_grad=True),
align_corners=False,
loss_decode=dict(
type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
ignore_index=8),
auxiliary_head=dict(
type='FCNHead',
in_channels=256,
in_index=2,
channels=256,
num_convs=1,
concat_input=False,
dropout_ratio=0.1,
num_classes=8,
norm_cfg=dict(type='BN', requires_grad=True),
align_corners=False,
loss_decode=dict(
type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.4),
ignore_index=8),
train_cfg=dict(),
test_cfg=dict(mode='whole'))
dataset_type = 'HuaweiDataset'
data_root = '/output/mmsegmentation/small_train'
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=(512, 512), ratio_range=(1.0, 1.0)),
dict(type='RandomFlip', flip_ratio=0.5),
dict(type='PhotoMetricDistortion'),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=255),
dict(type='DefaultFormatBundle'),
dict(type='Collect', keys=['img', 'gt_semantic_seg'])
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='MultiScaleFlipAug',
img_scale=(512, 512),
flip=False,
transforms=[
dict(type='Resize', keep_ratio=True),
dict(type='RandomFlip'),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='ImageToTensor', keys=['img']),
dict(type='Collect', keys=['img'])
])
]
test_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='MultiScaleFlipAug',
img_scale=(256, 256),
flip=False,
transforms=[
dict(type='Resize', keep_ratio=True),
dict(type='Resize', img_scale=(512, 512)),
dict(type='RandomFlip'),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='ImageToTensor', keys=['img']),
dict(type='Collect', keys=['img'])
])
]
data = dict(
samples_per_gpu=16,
workers_per_gpu=8,
train=dict(
type='HuaweiDataset',
data_root='/output/mmsegmentation/small_train',
img_dir='images',
ann_dir='labels',
pipeline=[
dict(type='LoadImageFromFile'),
dict(type='LoadAnnotations'),
dict(type='Resize', img_scale=(512, 512), ratio_range=(1.0, 1.0)),
dict(type='RandomFlip', flip_ratio=0.5),
dict(type='PhotoMetricDistortion'),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=255),
dict(type='DefaultFormatBundle'),
dict(type='Collect', keys=['img', 'gt_semantic_seg'])
],
split='splits/train.txt'),
val=dict(
type='HuaweiDataset',
data_root='/output/mmsegmentation/small_train',
img_dir='images',
ann_dir='anns',
pipeline=[
dict(type='LoadImageFromFile'),
dict(
type='MultiScaleFlipAug',
img_scale=(256, 256),
flip=False,
transforms=[
dict(type='Resize', keep_ratio=True),
dict(type='Resize', img_scale=(512, 512)),
dict(type='RandomFlip'),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='ImageToTensor', keys=['img']),
dict(type='Collect', keys=['img'])
])
],
split='splits/val.txt'),
test=dict(
type='HuaweiDataset',
data_root='/output/mmsegmentation/small_train',
img_dir='images',
ann_dir='anns',
pipeline=[
dict(type='LoadImageFromFile'),
dict(
type='MultiScaleFlipAug',
img_scale=(256, 256),
flip=False,
transforms=[
dict(type='Resize', keep_ratio=True),
dict(type='Resize', img_scale=(512, 512)),
dict(type='RandomFlip'),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='ImageToTensor', keys=['img']),
dict(type='Collect', keys=['img'])
])
],
split='splits/val.txt'))
log_config = dict(
interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])
dist_params = dict(backend='nccl')
log_level = 'INFO'
load_from = '/home/mmsegmentation/work_dirs/cd_train/iter_25000.pth'
resume_from = '/home/mmsegmentation/work_dirs/cd_train/iter_25000.pth'
workflow = [('train', 1)]
cudnn_benchmark = True
optimizer = dict(
type='AdamW',
lr=6e-05,
betas=(0.9, 0.999),
weight_decay=0.01,
paramwise_cfg=dict(
custom_keys=dict(
absolute_pos_embed=dict(decay_mult=0.0),
relative_position_bias_table=dict(decay_mult=0.0),
norm=dict(decay_mult=0.0))))
optimizer_config = dict()
lr_config = dict(
policy='poly',
warmup='linear',
warmup_iters=1500,
warmup_ratio=1e-06,
power=1.0,
min_lr=0.0,
by_epoch=False)
runner = dict(type='IterBasedRunner', max_iters=10000)
checkpoint_config = dict(by_epoch=False, interval=1000)
evaluation = dict(interval=1000, metric=['mIoU', 'mFscore'], pre_eval=True)
find_unused_parameters = True
work_dir = './work_dirs/cd_train'
gpu_ids = range(0, 1)
auto_resume = False
seed = 0
device = 'cuda'
训练是采用tools/train.py这个文件进行训练的,您能帮我解答为什么训练效果不好吗?万分感谢!
from vitae-transformer-remote-sensing.
@Liwiman 从你给的图看,大概率是标签没对齐,标签和类别的配置需要针对不同数据集进行专门设置,您可以参考以下issue
#9
#21
#23
from vitae-transformer-remote-sensing.
Related Issues (20)
- About Labels of Million-AID Dataset HOT 3
- 变化检测预训练权重问题 HOT 2
- About download the pretained model with change detection.
- KeyError: "EncoderDecoder: 'ViTAE_Window_NoShift_basic is not in the models registry'" HOT 7
- DIOR-R Benchmark question. HOT 1
- What are the differences between 'Your_ResNet' and MMCV's ResNet vb/vc/vd HOT 4
- Can't find the hrsc2016 in configs/_base_/datasets? HOT 1
- Reproduce the SeCo DOTA result. HOT 6
- use one image to test issues HOT 4
- Where is the train_labels_{}_{}.txt for scene recognition? HOT 2
- Semantic Segmentation: Potsdam 数据集复现性能差距有点大 HOT 15
- 模型注册问题 HOT 5
- 模型训练问题 HOT 3
- reproduce problem about swin-t in scene classification. HOT 1
- ann_file是什么格式的,怎么把八点法的labelTxt转成ann_file HOT 1
- label讀取的問題 HOT 1
- mmcv版本问题 HOT 1
- 模型预训练权重在哪下载 HOT 12
- questions about exp. of semantic seg. HOT 15
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from vitae-transformer-remote-sensing.