mirrors
/
mmsegmentation
mirror of https://github.com/open-mmlab/mmsegmentation.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Add more UNet-based medical segmentation benchmark (#324) 

* Add UNet as backbone and FCN PSPNet DeepLabV3 as decode_head benchmark on 4 retinal vessel segmentation datasets

* adjust README of UNet
pull/348/head
Junjun2016 2021-01-11 16:07:59 +08:00 committed by GitHub
parent da0af50812
commit 49515e7e3c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
16 changed files with 185 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

50
configs/_base_/models/deeplabv3_unet_s5-d16.py 100644
View File

@ -0,0 +1,50 @@
# model settings
norm_cfg = dict(type='SyncBN', requires_grad=True)
model = dict(
type='EncoderDecoder',
pretrained=None,
backbone=dict(
type='UNet',
in_channels=3,
base_channels=64,
num_stages=5,
strides=(1, 1, 1, 1, 1),
enc_num_convs=(2, 2, 2, 2, 2),
dec_num_convs=(2, 2, 2, 2),
downsamples=(True, True, True, True),
enc_dilations=(1, 1, 1, 1, 1),
dec_dilations=(1, 1, 1, 1),
with_cp=False,
conv_cfg=None,
norm_cfg=norm_cfg,
act_cfg=dict(type='ReLU'),
upsample_cfg=dict(type='InterpConv'),
norm_eval=False),
decode_head=dict(
type='ASPPHead',
in_channels=64,
in_index=4,
channels=16,
dilations=(1, 12, 24, 36),
dropout_ratio=0.1,
num_classes=2,
norm_cfg=norm_cfg,
align_corners=False,
loss_decode=dict(
type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
auxiliary_head=dict(
type='FCNHead',
in_channels=128,
in_index=3,
channels=64,
num_convs=1,
concat_input=False,
dropout_ratio=0.1,
num_classes=2,
norm_cfg=norm_cfg,
align_corners=False,
loss_decode=dict(
type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.4)))
# model training and testing settings
train_cfg = dict()
test_cfg = dict(mode='slide', crop_size=256, stride=170)

0
configs/_base_/models/unet_s5-d16.py → configs/_base_/models/fcn_unet_s5-d16.py
View File

50
configs/_base_/models/pspnet_unet_s5-d16.py 100644
View File

@ -0,0 +1,50 @@
# model settings
norm_cfg = dict(type='SyncBN', requires_grad=True)
model = dict(
type='EncoderDecoder',
pretrained=None,
backbone=dict(
type='UNet',
in_channels=3,
base_channels=64,
num_stages=5,
strides=(1, 1, 1, 1, 1),
enc_num_convs=(2, 2, 2, 2, 2),
dec_num_convs=(2, 2, 2, 2),
downsamples=(True, True, True, True),
enc_dilations=(1, 1, 1, 1, 1),
dec_dilations=(1, 1, 1, 1),
with_cp=False,
conv_cfg=None,
norm_cfg=norm_cfg,
act_cfg=dict(type='ReLU'),
upsample_cfg=dict(type='InterpConv'),
norm_eval=False),
decode_head=dict(
type='PSPHead',
in_channels=64,
in_index=4,
channels=16,
pool_scales=(1, 2, 3, 6),
dropout_ratio=0.1,
num_classes=2,
norm_cfg=norm_cfg,
align_corners=False,
loss_decode=dict(
type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
auxiliary_head=dict(
type='FCNHead',
in_channels=128,
in_index=3,
channels=64,
num_convs=1,
concat_input=False,
dropout_ratio=0.1,
num_classes=2,
norm_cfg=norm_cfg,
align_corners=False,
loss_decode=dict(
type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.4)))
# model training and testing settings
train_cfg = dict()
test_cfg = dict(mode='slide', crop_size=256, stride=170)

37
configs/unet/README.md
View File

@ -17,9 +17,34 @@
## Results and models
| Backbone | Head | Dataset | Image Size | Crop Size | Stride | Lr schd | Mem (GB) | Inf time (fps) | Dice | download |
|--------|----------|----------|----------|-----------|--------:|----------|----------------|------:|--------------:|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| UNet-S5-D16 | FCN | DRIVE | 584x565 | 64x64 | 42x42 | 40000 | 0.680 | - | 78.67 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/unet_s5-d16_64x64_40k_drive/unet_s5-d16_64x64_40k_drive_20201223_191051-9cd163b8.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/unet_s5-d16_64x64_40k_drive/unet_s5-d16_64x64_40k_drive-20201223_191051.log.json) |
| UNet-S5-D16 | FCN | STARE | 605x700 | 128x128 | 85x85 | 40000 | 0.968 | - | 81.02 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/unet_s5-d16_128x128_40k_stare/unet_s5-d16_128x128_40k_stare_20201223_191051-e5439846.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/unet_s5-d16_128x128_40k_stare/unet_s5-d16_128x128_40k_stare-20201223_191051.log.json) |
| UNet-S5-D16 | FCN | CHASE_DB1 | 960x999 | 128x128 | 85x85 | 40000 | 0.968 | - | 80.24 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/unet_s5-d16_128x128_40k_chase_db1/unet_s5-d16_128x128_40k_chase_db1_20201223_191051-8b16ca0b.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/unet_s5-d16_128x128_40k_chase_db1/unet_s5-d16_128x128_40k_chase_db1-20201223_191051.log.json) |
| UNet-S5-D16 | FCN | HRF | 2336x3504 | 256x256 | 170x170 | 40000 | 2.525 | - | 79.45 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/unet_s5-d16_256x256_40k_hrf/unet_s5-d16_256x256_40k_hrf_20201223_173724-d89cf1ed.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/unet_s5-d16_256x256_40k_hrf/unet_s5-d16_256x256_40k_hrf-20201223_173724.log.json) |
### DRIVE
| Backbone | Head | Image Size | Crop Size | Stride | Lr schd | Mem (GB) | Inf time (fps) | Dice | download |
|--------|----------|----------|-----------|--------:|----------|----------------|------:|--------------:|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| UNet-S5-D16 | FCN | 584x565 | 64x64 | 42x42 | 40000 | 0.680 | - | 78.67 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/fcn_unet_s5-d16_64x64_40k_drive/fcn_unet_s5-d16_64x64_40k_drive_20201223_191051-26cee593.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/fcn_unet_s5-d16_64x64_40k_drive/fcn_unet_s5-d16_64x64_40k_drive-20201223_191051.log.json) |
| UNet-S5-D16 | PSPNet | 584x565 | 64x64 | 42x42 | 40000 | 0.599 | - | 78.62 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/pspnet_unet_s5-d16_64x64_40k_drive/pspnet_unet_s5-d16_64x64_40k_drive_20201227_181818-aac73387.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/pspnet_unet_s5-d16_64x64_40k_drive/pspnet_unet_s5-d16_64x64_40k_drive-20201227_181818.log.json) |
| UNet-S5-D16 | DeepLabV3 | 584x565 | 64x64 | 42x42 | 40000 | 0.596 | - | 78.69 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/deeplabv3_unet_s5-d16_64x64_40k_drive/deeplabv3_unet_s5-d16_64x64_40k_drive_20201226_094047-0671ff20.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/deeplabv3_unet_s5-d16_64x64_40k_drive/deeplabv3_unet_s5-d16_64x64_40k_drive-20201226_094047.log.json) |
### STARE
| Backbone | Head | Image Size | Crop Size | Stride | Lr schd | Mem (GB) | Inf time (fps) | Dice | download |
|--------|----------|----------|-----------|--------:|----------|----------------|------:|--------------:|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| UNet-S5-D16 | FCN | 605x700 | 128x128 | 85x85 | 40000 | 0.968 | - | 81.02 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/fcn_unet_s5-d16_128x128_40k_stare/fcn_unet_s5-d16_128x128_40k_stare_20201223_191051-6ea7cfda.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/fcn_unet_s5-d16_128x128_40k_stare/fcn_unet_s5-d16_128x128_40k_stare-20201223_191051.log.json) |
| UNet-S5-D16 | PSPNet | 605x700 | 128x128 | 85x85 | 40000 | 0.982 | - | 81.22 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/pspnet_unet_s5-d16_128x128_40k_stare/pspnet_unet_s5-d16_128x128_40k_stare_20201227_181818-3c2923c4.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/pspnet_unet_s5-d16_128x128_40k_stare/pspnet_unet_s5-d16_128x128_40k_stare-20201227_181818.log.json) |
| UNet-S5-D16 | DeepLabV3 | 605x700 | 128x128 | 85x85 | 40000 | 0.999 | - | 80.93 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/deeplabv3_unet_s5-d16_128x128_40k_stare/deeplabv3_unet_s5-d16_128x128_40k_stare_20201226_094047-93dcb93c.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/deeplabv3_unet_s5-d16_128x128_40k_stare/deeplabv3_unet_s5-d16_128x128_40k_stare-20201226_094047.log.json) |
### CHASE_DB1
| Backbone | Head | Image Size | Crop Size | Stride | Lr schd | Mem (GB) | Inf time (fps) | Dice | download |
|--------|----------|----------|-----------|--------:|----------|----------------|------:|--------------:|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| UNet-S5-D16 | FCN | 960x999 | 128x128 | 85x85 | 40000 | 0.968 | - | 80.24 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/fcn_unet_s5-d16_128x128_40k_chase_db1/fcn_unet_s5-d16_128x128_40k_chase_db1_20201223_191051-95852f45.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/fcn_unet_s5-d16_128x128_40k_chase_db1/fcn_unet_s5-d16_128x128_40k_chase_db1-20201223_191051.log.json) |
| UNet-S5-D16 | PSPNet | 960x999 | 128x128 | 85x85 | 40000 | 0.982 | - | 80.36 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/pspnet_unet_s5-d16_128x128_40k_chase_db1/pspnet_unet_s5-d16_128x128_40k_chase_db1_20201227_181818-68d4e609.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/pspnet_unet_s5-d16_128x128_40k_chase_db1/pspnet_unet_s5-d16_128x128_40k_chase_db1-20201227_181818.log.json) |
| UNet-S5-D16 | DeepLabV3 | 960x999 | 128x128 | 85x85 | 40000 | 0.999 | - | 80.47 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/deeplabv3_unet_s5-d16_128x128_40k_chase_db1/deeplabv3_unet_s5-d16_128x128_40k_chase_db1_20201226_094047-4c5aefa3.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/deeplabv3_unet_s5-d16_128x128_40k_chase_db1/deeplabv3_unet_s5-d16_128x128_40k_chase_db1-20201226_094047.log.json) |
### HRF
| Backbone | Head | Image Size | Crop Size | Stride | Lr schd | Mem (GB) | Inf time (fps) | Dice | download |
|--------|----------|----------|-----------|--------:|----------|----------------|------:|--------------:|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| UNet-S5-D16 | FCN | 2336x3504 | 256x256 | 170x170 | 40000 | 2.525 | - | 79.45 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/fcn_unet_s5-d16_256x256_40k_hrf/fcn_unet_s5-d16_256x256_40k_hrf_20201223_173724-df3ec8c4.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/fcn_unet_s5-d16_256x256_40k_hrf/fcn_unet_s5-d16_256x256_40k_hrf-20201223_173724.log.json) |
| UNet-S5-D16 | PSPNet | 2336x3504 | 256x256 | 170x170 | 40000 | 2.588 | - | 80.07 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/pspnet_unet_s5-d16_256x256_40k_hrf/pspnet_unet_s5-d16_256x256_40k_hrf_20201227_181818-fdb7e29b.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/pspnet_unet_s5-d16_256x256_40k_hrf/pspnet_unet_s5-d16_256x256_40k_hrf-20201227_181818.log.json) |
| UNet-S5-D16 | DeepLabV3 | 2336x3504 | 256x256 | 170x170 | 40000 | 2.604 | - | 80.21 | [model](https://download.openmmlab.com/mmsegmentation/v0.5/unet/deeplabv3_unet_s5-d16_256x256_40k_hrf/deeplabv3_unet_s5-d16_256x256_40k_hrf_20201226_094047-3a1fdf85.pth) | [log](https://download.openmmlab.com/mmsegmentation/v0.5/unet/deeplabv3_unet_s5-d16_256x256_40k_hrf/deeplabv3_unet_s5-d16_256x256_40k_hrf-20201226_094047.log.json) |

7
configs/unet/deeplabv3_unet_s5-d16_128x128_40k_chase_db1.py 100644
View File

@ -0,0 +1,7 @@
_base_ = [
'../_base_/models/deeplabv3_unet_s5-d16.py',
'../_base_/datasets/chase_db1.py', '../_base_/default_runtime.py',
'../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(128, 128), stride=(85, 85))
evaluation = dict(metric='mDice')

6
configs/unet/deeplabv3_unet_s5-d16_128x128_40k_stare.py 100644
View File

@ -0,0 +1,6 @@
_base_ = [
'../_base_/models/deeplabv3_unet_s5-d16.py', '../_base_/datasets/stare.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(128, 128), stride=(85, 85))
evaluation = dict(metric='mDice')

6
configs/unet/deeplabv3_unet_s5-d16_256x256_40k_hrf.py 100644
View File

@ -0,0 +1,6 @@
_base_ = [
'../_base_/models/deeplabv3_unet_s5-d16.py', '../_base_/datasets/hrf.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(256, 256), stride=(170, 170))
evaluation = dict(metric='mDice')

6
configs/unet/deeplabv3_unet_s5-d16_64x64_40k_drive.py 100644
View File

@ -0,0 +1,6 @@
_base_ = [
'../_base_/models/deeplabv3_unet_s5-d16.py', '../_base_/datasets/drive.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(64, 64), stride=(42, 42))
evaluation = dict(metric='mDice')

6
configs/unet/fcn_unet_s5-d16_128x128_40k_chase_db1.py 100644
View File

@ -0,0 +1,6 @@
_base_ = [
'../_base_/models/fcn_unet_s5-d16.py', '../_base_/datasets/chase_db1.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(128, 128), stride=(85, 85))
evaluation = dict(metric='mDice')

2
configs/unet/unet_s5-d16_128x128_40k_chase_db1.py → configs/unet/fcn_unet_s5-d16_128x128_40k_stare.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../_base_/models/unet_s5-d16.py', '../_base_/datasets/chase_db1.py',
'../_base_/models/fcn_unet_s5-d16.py', '../_base_/datasets/stare.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(128, 128), stride=(85, 85))

2
configs/unet/unet_s5-d16_256x256_40k_hrf.py → configs/unet/fcn_unet_s5-d16_256x256_40k_hrf.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../_base_/models/unet_s5-d16.py', '../_base_/datasets/hrf.py',
'../_base_/models/fcn_unet_s5-d16.py', '../_base_/datasets/hrf.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(256, 256), stride=(170, 170))

2
configs/unet/unet_s5-d16_64x64_40k_drive.py → configs/unet/fcn_unet_s5-d16_64x64_40k_drive.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../_base_/models/unet_s5-d16.py', '../_base_/datasets/drive.py',
'../_base_/models/fcn_unet_s5-d16.py', '../_base_/datasets/drive.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(64, 64), stride=(42, 42))

7
configs/unet/pspnet_unet_s5-d16_128x128_40k_chase_db1.py 100644
View File

@ -0,0 +1,7 @@
_base_ = [
'../_base_/models/pspnet_unet_s5-d16.py',
'../_base_/datasets/chase_db1.py', '../_base_/default_runtime.py',
'../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(128, 128), stride=(85, 85))
evaluation = dict(metric='mDice')

2
configs/unet/unet_s5-d16_128x128_40k_stare.py → configs/unet/pspnet_unet_s5-d16_128x128_40k_stare.py
View File

@ -1,5 +1,5 @@
_base_ = [
'../_base_/models/unet_s5-d16.py', '../_base_/datasets/stare.py',
'../_base_/models/pspnet_unet_s5-d16.py', '../_base_/datasets/stare.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(128, 128), stride=(85, 85))

6
configs/unet/pspnet_unet_s5-d16_256x256_40k_hrf.py 100644
View File

@ -0,0 +1,6 @@
_base_ = [
'../_base_/models/pspnet_unet_s5-d16.py', '../_base_/datasets/hrf.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(256, 256), stride=(170, 170))
evaluation = dict(metric='mDice')

6
configs/unet/pspnet_unet_s5-d16_64x64_40k_drive.py 100644
View File

@ -0,0 +1,6 @@
_base_ = [
'../_base_/models/pspnet_unet_s5-d16.py', '../_base_/datasets/drive.py',
'../_base_/default_runtime.py', '../_base_/schedules/schedule_40k.py'
]
test_cfg = dict(crop_size=(64, 64), stride=(42, 42))
evaluation = dict(metric='mDice')