CodeCamp #140 [Feature] Add synapse dataset and data augmentation in dev-1.x. #2372

Dominic23331 · 2022-12-03T02:01:34Z

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

Support synapse and data augmentation
dataset link: https://www.synapse.org/#!Synapse:syn3193805/wiki/
TransUnet use this dataset, paper link: https://arxiv.org/pdf/2102.04306.pdf

Modification

Add synapse dataset loader.
Add a python script to transform synapse dataset format to mmseg dataset format.
Add synapse dataset augmentation.

Update 2022-12-13

The result when using 13 classes:

        classes=('background', 'spleen', 'right_kidney', 'left_kidney',
                 'gallbladder', 'esophagus', 'liver', 'stomach', 'aorta',
                 'inferior_vena_cava', 'portal_vein_and_splenic_vein',
                 'pancreas', 'right_adrenal_gland', 'left_adrenal_gland'),
        palette=[[0, 0, 0], [255, 127, 127], [224, 231, 161], [138, 204, 132],
                 [64, 172, 136], [126, 152, 187], [140, 110, 160],
                 [247, 88, 240], [202, 172, 161], [237, 213, 149],
                 [139, 182, 139], [111, 192, 185], [82, 107, 163],
                 [89, 54, 156]])

CLAassistant · 2022-12-03T02:01:38Z

All committers have signed the CLA.

MengzhangLI · 2022-12-03T04:26:40Z

Hi, thanks for your nice PR. We would review it asap.

Best wishes,

codecov · 2022-12-03T04:36:14Z

Codecov Report

Base: 83.33% // Head: 83.40% // Increases project coverage by +0.06% 🎉

Coverage data is based on head (956d75f) compared to base (534b27b).
Patch coverage: 94.11% of modified lines in pull request are covered.

Additional details and impacted files

@@             Coverage Diff             @@
##           dev-1.x    #2372      +/-   ##
===========================================
+ Coverage    83.33%   83.40%   +0.06%     
===========================================
  Files          143      144       +1     
  Lines         8127     8178      +51     
  Branches      1211     1219       +8     
===========================================
+ Hits          6773     6821      +48     
- Misses        1165     1168       +3     
  Partials       189      189

Flag	Coverage Δ
unittests	`83.40% <94.11%> (+0.06%)`	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
mmseg/datasets/transforms/__init__.py	`100.00% <ø> (ø)`
mmseg/utils/__init__.py	`100.00% <ø> (ø)`
mmseg/datasets/transforms/transforms.py	`96.09% <92.30%> (-0.34%)`	⬇️
mmseg/datasets/__init__.py	`100.00% <100.00%> (ø)`
mmseg/datasets/synapse.py	`100.00% <100.00%> (ø)`
mmseg/utils/class_names.py	`86.76% <100.00%> (+0.82%)`	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

tests/test_datasets/test_dataset.py

mmseg/datasets/synapse.py

MengzhangLI · 2022-12-09T13:33:40Z

tools/dataset_converters/synapse.py

+    mkdir_or_exist(osp.join(save_path, 'img_dir'))
+    mkdir_or_exist(osp.join(save_path, 'ann_dir'))
+
+    if osp.exists(osp.join(dataset_path, 'train.txt')) \


Could you please provide train.txt and val.txt which makes the same with TransUNet train/val split? In this default setting, all datasets are handled into training set.

The dataset split could be found here: https://github.com/Beckschen/TransUNet/tree/main/lists/lists_Synapse.

MengzhangLI · 2022-12-09T14:01:56Z

mmseg/datasets/synapse.py

+
+    def __init__(self,
+                 img_suffix='.jpg',
+                 seg_map_suffix='.png',


In default setting, the label segmentation map would be .jpg rather than .png. link of code.

MengzhangLI · 2022-12-12T11:05:59Z

configs/_base_/datasets/synapse.py

@@ -0,0 +1,41 @@
+dataset_type = 'SynapseDataset'


The setting of this synapse dataset config may be aligned with TransUNet default setting: https://github.com/Beckschen/TransUNet/blob/main/train.py#L19-L33.

MengzhangLI · 2022-12-13T14:58:54Z

mmseg/datasets/synapse.py

+        classes=('background', 'spleen', 'right_kidney', 'left_kidney',
+                 'gallbladder', 'esophagus', 'liver', 'stomach', 'aorta',
+                 'inferior_vena_cava', 'portal_vein_and_splenic_vein',
+                 'pancreas', 'right_adrenal_gland', 'left_adrenal_gland'),
+        palette=[[0, 0, 0], [255, 127, 127], [224, 231, 161], [138, 204, 132],
+                 [64, 172, 136], [126, 152, 187], [140, 110, 160],
+                 [247, 88, 240], [202, 172, 161], [237, 213, 149],
+                 [139, 182, 139], [111, 192, 185], [82, 107, 163],
+                 [89, 54, 156]])


Perhaps we should keep label number the same with TransUNet, i.e., only 'spleen', 'right_kidney', 'left_kidney', 'gallbladder', 'liver', 'stomach', 'aorta', 'pancreas' 8 classes are handled. Because it has been introduced as benchmark in many medical image segmentation paper.

MengzhangLI · 2022-12-13T15:16:57Z

configs/_base_/datasets/synapse.py

+        pipeline=test_pipeline))
+test_dataloader = val_dataloader
+
+val_evaluator = dict(type='IoUMetric', iou_metrics=['mIoU'])


In evaluation, the setting of TransUNet and its following works all use DSC based on 3D scan rather than 2D slice which is MMSegmentation default setting. We may use evaluation based on 3D scans to make users more convenient when using Synapse dataset in MMSegmentation.

MengzhangLI · 2022-12-20T12:36:47Z

First, we may use dataset from official repo, which can be downloaded by public http link, rather than TransUNet unofficial processed data.

Then we should ensure the our dataset conversion could handle dataset the same with TransUNet, Convert them to numpy format, clip the images within [-125, 275], normalize each 3D image to [0, 1], and extract 2D slices from 3D volume for training cases while keeping the 3D volume in h5 format for testing cases. The shape (H, W) and pixel value of each image should be the same.

Next, we should check out normalization of TransUNet pretrained model, use its parameters in our backbone config, like here.

Thanks for your nice PR again!

MengzhangLI · 2022-12-20T12:43:24Z

Could you please add doc for how to download the original data and how to convert it in dataset_prepare.md?

MengzhangLI · 2022-12-21T07:53:35Z

tools/dataset_converters/synapse.py

+        label_3d = read_nii_file(
+            osp.join(dataset_path, 'label', 'label' + idx + '.nii.gz'))
+
+        img_3d = np.clip(img_3d, -125, 275)


Add comment above this line, give a link: https://github.com/Beckschen/TransUNet/tree/main/datasets to tell users why we clip (-125, 275). Thx.

MengzhangLI · 2022-12-23T07:42:58Z

This PR is wired to be closed automatically when I git push some configs for model training, @Dominic23331 could you please make a new PR agian? Sorry for my wrong operation!

Best,

…on in dev-1.x. (#2432) ## Motivation Add Synapse dataset in MMSegmentation. Old PR: #2372.

mm-assistant bot assigned xiexinch Dec 3, 2022

MengzhangLI self-assigned this Dec 4, 2022

MengzhangLI changed the title ~~[feature] Add synapse dataset and data augmentation~~ [Feature] Add synapse dataset and data augmentation in dev-1.x. Dec 5, 2022

MengzhangLI added the 1.x Related issue of 1.x version label Dec 5, 2022

MengzhangLI changed the title ~~[Feature] Add synapse dataset and data augmentation in dev-1.x.~~ CodeCamp #140 [Feature] Add synapse dataset and data augmentation in dev-1.x. Dec 5, 2022

MengzhangLI reviewed Dec 8, 2022

View reviewed changes

tests/test_datasets/test_dataset.py Outdated Show resolved Hide resolved

MengzhangLI reviewed Dec 8, 2022

View reviewed changes

mmseg/datasets/synapse.py Outdated Show resolved Hide resolved

MengzhangLI reviewed Dec 9, 2022

View reviewed changes

MengzhangLI reviewed Dec 12, 2022

View reviewed changes

MengzhangLI reviewed Dec 13, 2022

View reviewed changes

MeowZheng added the medical label Dec 15, 2022

MengzhangLI reviewed Dec 21, 2022

View reviewed changes

MengzhangLI closed this Dec 23, 2022

MengzhangLI mentioned this pull request Dec 23, 2022

CodeCamp #140 [New] [Feature] Add synapse dataset and data augmentation in dev-1.x. #2432

Merged

MeowZheng pushed a commit that referenced this pull request Jan 6, 2023

CodeCamp #140 [New] [Feature] Add synapse dataset and data augmentati…

2d67e51

…on in dev-1.x. (#2432) ## Motivation Add Synapse dataset in MMSegmentation. Old PR: #2372.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

CodeCamp #140 [Feature] Add synapse dataset and data augmentation in dev-1.x. #2372

CodeCamp #140 [Feature] Add synapse dataset and data augmentation in dev-1.x. #2372

Dominic23331 commented Dec 3, 2022 •

edited by MengzhangLI

Loading

CLAassistant commented Dec 3, 2022 •

edited

Loading

MengzhangLI commented Dec 3, 2022

codecov bot commented Dec 3, 2022 •

edited

Loading

MengzhangLI Dec 9, 2022

MengzhangLI Dec 12, 2022

MengzhangLI Dec 9, 2022

MengzhangLI Dec 12, 2022

MengzhangLI Dec 13, 2022

MengzhangLI Dec 13, 2022 •

edited

Loading

MengzhangLI commented Dec 20, 2022

MengzhangLI commented Dec 20, 2022

MengzhangLI Dec 21, 2022

MengzhangLI commented Dec 23, 2022

CodeCamp #140 [Feature] Add synapse dataset and data augmentation in dev-1.x. #2372

CodeCamp #140 [Feature] Add synapse dataset and data augmentation in dev-1.x. #2372

Conversation

Dominic23331 commented Dec 3, 2022 • edited by MengzhangLI Loading

Motivation

Modification

Update 2022-12-13

CLAassistant commented Dec 3, 2022 • edited Loading

MengzhangLI commented Dec 3, 2022

codecov bot commented Dec 3, 2022 • edited Loading

Codecov Report

MengzhangLI Dec 9, 2022

Choose a reason for hiding this comment

MengzhangLI Dec 12, 2022

Choose a reason for hiding this comment

MengzhangLI Dec 9, 2022

Choose a reason for hiding this comment

MengzhangLI Dec 12, 2022

Choose a reason for hiding this comment

MengzhangLI Dec 13, 2022

Choose a reason for hiding this comment

MengzhangLI Dec 13, 2022 • edited Loading

Choose a reason for hiding this comment

MengzhangLI commented Dec 20, 2022

MengzhangLI commented Dec 20, 2022

MengzhangLI Dec 21, 2022

Choose a reason for hiding this comment

MengzhangLI commented Dec 23, 2022

Dominic23331 commented Dec 3, 2022 •

edited by MengzhangLI

Loading

CLAassistant commented Dec 3, 2022 •

edited

Loading

codecov bot commented Dec 3, 2022 •

edited

Loading

MengzhangLI Dec 13, 2022 •

edited

Loading