Add support for Pascal Context 59 classes #459

clownrat6 · 2021-04-08T02:05:40Z

Create PascalContextDataset59 class in mmseg/datasets/pascal_context.py;
Set reduce_zero_label=True for train_pipeline and PascalContextDataset59;
Add some configs for Pascal-Context 59 classes training and testing;

All evaluations are conducted after PR #471.

This table aims at checking the correctness of pascal context 59 dataset support.

model	test repo	num classes	Lr schd	mIoU	mIoU(ms+flip)
HRNetV2-W48(author)	mmsegmentation	59	-	51.7	53.92
HRNetV2-W48(author)	mmsegmentation	60	-	46.18	48.02
HRNetV2-W48(author)	raw	59	-	52.25	54.1
HRNetV2-W48(author)	raw	60	-	46.96	48.3
HRNetV2-W48(new)	mmsegmentation	59	40000	50.37	52.88
HRNetV2-W48(new)	mmsegmentation	60	40000	45.55	47.7
HRNetV2-W48(new)	mmsegmentation	59	80000	51.12	53.56
HRNetV2-W48(new	mmsegmentation	60	80000	45.79	47.86
HRNetV2-W48(old)	mmsegmentation	60	40000	45.14	47.42

This table is the training and evaluation collection.

model	num classes	Lr schd	mIoU	mIoU(ms+flip)
DeepLabV3+_R-101-D8	59	40000	52.86	54.54
DeepLabV3+_R-101-D8	59	80000	53.2	54.67
DeepLabV3_R-101-D8	59	40000	52.61	54.28
DeepLabV3_R-101-D8	59	80000	52.46	54.09
PSPNet_R-101-D8	59	40000	52.02	53.54
PSPNet_R-101-D8	59	80000	52.47	53.99
HRNetV2-W48	59	40000	50.33	52.83
HRNetV2-W48	59	80000	51.12	53.56
FCN_R-101-D8	59	40000	48.42	50.4
FCN_R-101-D8	59	80000	49.35	51.38

How to use HRNetV2-W48 official weights to test on our repo?

ps: testing on pascal context dataset

Download HRNetV2-W48 official weights 59 classes and 69 classes. (You'd better download them in master branch of official repo)
Then, using the convert script to convert official weights to pytorch style:

python tools/scripts/convert_hrnet.py [src_path] [dst_path]

In order to match the inference process, modify codes according to this commit.
Then, using hrnet config to test model on pascal context dataset:

python -u tools/test.py [config path] [converted official weights path] --eval mIoU

CLAassistant · 2021-04-08T02:05:43Z

All committers have signed the CLA.

mmseg/datasets/pascal_context.py

configs/_base_/datasets/pascal_context_59.py

codecov · 2021-04-08T02:19:54Z

Codecov Report

Merging #459 (a768991) into master (1483720) will decrease coverage by 0.02%.
The diff coverage is 77.77%.

@@            Coverage Diff             @@
##           master     #459      +/-   ##
==========================================
- Coverage   86.51%   86.48%   -0.03%     
==========================================
  Files          97       97              
  Lines        4967     4974       +7     
  Branches      806      807       +1     
==========================================
+ Hits         4297     4302       +5     
- Misses        517      519       +2     
  Partials      153      153

Flag	Coverage Δ
unittests	`86.48% <77.77%> (-0.03%)`	⬇️

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

Impacted Files	Coverage Δ
mmseg/datasets/pascal_context.py	`76.47% <75.00%> (-3.53%)`	⬇️
mmseg/datasets/__init__.py	`100.00% <100.00%> (ø)`

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1483720...a768991. Read the comment docs.

configs/_base_/datasets/pascal_context_59.py

* Create PascalContextDataset59 class in mmseg/datasets/pascal_context.py; * Set reduce_zero_label=True for train_pipeline and PascalContextDataset59; * Add some configs for Pascal-Context 59 classes training and testing;

…)=0";

…)=0"; * Modify the calculation of IoU;

xvjiarui · 2021-04-09T18:27:54Z

mmseg/core/evaluation/metrics.py

-        intersect.float(), bins=(num_classes), min=0, max=num_classes)
+        intersect.float(), bins=num_classes, min=0, max=num_classes - 1)


We may need to test whether the results are consistent for other models.
We could make another PR for this modification, we may also include impact range in the new PR.

We may merge master and resolve conflicts.

…training based on PascalContextDataset59; Add some ignore items in ".gitignore";

… of pspnet config file;

.gitignore

configs/_base_/schedules/schedule_20k_epochwise.py

* Add weighs link, config link, log link and evaluation results about PascalContextDataset59 in README.md

…nsmit port when you transmit data to distributed machine.

xvjiarui · 2021-04-19T16:54:25Z

Please merge master to resolve the conflicts.

xvjiarui · 2021-04-19T16:54:34Z

We may delete unused configs.

tools/test.py

configs/hrnet/fcn_hr18s_480x480_40k_pascal_context_59.py

* Remove "-p|--port" command argument;

* Add support for Pascal Context 59 classes (open-mmlab#459) * Create PascalContextDataset59 class in mmseg/datasets/pascal_context.py; * Set reduce_zero_label=True for train_pipeline and PascalContextDataset59; * Add some configs for Pascal-Context 59 classes training and testing; * Try to solve the problem about "fence(IoU)=nan grass(IoU)=0"; * Continue(1): Try to solve the problem about "fence(IoU)=nan grass(IoU)=0"; * ignore files and folders named tempxxx; * Continue(2): Try to solve the problem about "fence(IoU)=nan grass(IoU)=0"; * Modify the calculation of IoU; * Modify the CLASSES order of PascalContextDataset; * Add "fcn", "deeplabv3", "deeplabv3+", "pspnet" config file for model training based on PascalContextDataset59; Add some ignore items in ".gitignore"; * fix the bug "test_cfg specified in both outer field and model field " of pspnet config file; * * Clean unnecessary codes; * Add weighs link, config link, log link and evaluation results about PascalContextDataset59 in README.md * Add command line argument: "-p | --port", this arg can change the transmit port when you transmit data to distributed machine. * * Remove rebundant config files; * Remove "-p|--port" command argument; Co-authored-by: Jiarui XU <xvjiarui0826@gmail.com>

xvjiarui reviewed Apr 8, 2021

View reviewed changes

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

xvjiarui reviewed Apr 8, 2021

View reviewed changes

configs/_base_/datasets/pascal_context_59.py Outdated Show resolved Hide resolved

xvjiarui reviewed Apr 8, 2021

View reviewed changes

configs/_base_/datasets/pascal_context_59.py Show resolved Hide resolved

clownrat6 force-pushed the debug branch from 1dabc1f to 1e1af3b Compare April 8, 2021 03:16

clownrat6 changed the title ~~add some pascal context 59 classes configs~~ Add support for Pascal Context 59 classes Apr 8, 2021

Add support for Pascal Context 59 classes (open-mmlab#459)

b081822

* Create PascalContextDataset59 class in mmseg/datasets/pascal_context.py; * Set reduce_zero_label=True for train_pipeline and PascalContextDataset59; * Add some configs for Pascal-Context 59 classes training and testing;

clownrat6 force-pushed the debug branch from 1e1af3b to b081822 Compare April 8, 2021 04:06

Try to solve the problem about "fence(IoU)=nan grass(IoU)=0";

8371ba5

clownrat6 force-pushed the debug branch from f898336 to 8371ba5 Compare April 8, 2021 08:09

sennnnn added 4 commits April 8, 2021 20:13

Continue(1): Try to solve the problem about "fence(IoU)=nan grass(IoU…

f5ab1f9

…)=0";

ignore files and folders named tempxxx;

3379dd3

Continue(2): Try to solve the problem about "fence(IoU)=nan grass(IoU…

c2be19a

…)=0"; * Modify the calculation of IoU;

Modify the CLASSES order of PascalContextDataset;

42ea672

xvjiarui reviewed Apr 9, 2021

View reviewed changes

sennnnn added 2 commits April 16, 2021 11:00

Add "fcn", "deeplabv3", "deeplabv3+", "pspnet" config file for model …

2de91d7

…training based on PascalContextDataset59; Add some ignore items in ".gitignore";

fix the bug "test_cfg specified in both outer field and model field "…

9fb99b4

… of pspnet config file;

clownrat6 force-pushed the debug branch from b6f7690 to 9fb99b4 Compare April 16, 2021 05:49

xvjiarui reviewed Apr 16, 2021

View reviewed changes

.gitignore Outdated Show resolved Hide resolved

xvjiarui reviewed Apr 16, 2021

View reviewed changes

configs/_base_/schedules/schedule_20k_epochwise.py Outdated Show resolved Hide resolved

sennnnn added 2 commits April 19, 2021 16:57

* Clean unnecessary codes;

efa651e

* Add weighs link, config link, log link and evaluation results about PascalContextDataset59 in README.md

Add command line argument: "-p | --port", this arg can change the tra…

d463783

…nsmit port when you transmit data to distributed machine.

xvjiarui reviewed Apr 19, 2021

View reviewed changes

tools/test.py Outdated Show resolved Hide resolved

clownrat6 force-pushed the debug branch 2 times, most recently from 1c5a7eb to d463783 Compare April 20, 2021 02:11

xvjiarui reviewed Apr 20, 2021

View reviewed changes

configs/hrnet/fcn_hr18s_480x480_40k_pascal_context_59.py Show resolved Hide resolved

merge master

f65e346

* Remove rebundant config files;

a768991

* Remove "-p|--port" command argument;

xvjiarui approved these changes Apr 20, 2021

View reviewed changes

xvjiarui merged commit 83d312e into open-mmlab:master Apr 20, 2021

sibozhang pushed a commit to sibozhang/mmsegmentation that referenced this pull request Mar 22, 2024

fix bug when users do not install mmdet (open-mmlab#459)

da2ddf1

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add support for Pascal Context 59 classes #459

Add support for Pascal Context 59 classes #459

clownrat6 commented Apr 8, 2021 •

edited

Loading

CLAassistant commented Apr 8, 2021 •

edited

Loading

codecov bot commented Apr 8, 2021 •

edited

Loading

xvjiarui Apr 9, 2021

xvjiarui Apr 16, 2021

xvjiarui commented Apr 19, 2021

xvjiarui commented Apr 19, 2021

		intersect.float(), bins=(num_classes), min=0, max=num_classes)
		intersect.float(), bins=num_classes, min=0, max=num_classes - 1)

Add support for Pascal Context 59 classes #459

Add support for Pascal Context 59 classes #459

Conversation

clownrat6 commented Apr 8, 2021 • edited Loading

How to use HRNetV2-W48 official weights to test on our repo?

CLAassistant commented Apr 8, 2021 • edited Loading

codecov bot commented Apr 8, 2021 • edited Loading

Codecov Report

xvjiarui Apr 9, 2021

Choose a reason for hiding this comment

xvjiarui Apr 16, 2021

Choose a reason for hiding this comment

xvjiarui commented Apr 19, 2021

xvjiarui commented Apr 19, 2021

clownrat6 commented Apr 8, 2021 •

edited

Loading

CLAassistant commented Apr 8, 2021 •

edited

Loading

codecov bot commented Apr 8, 2021 •

edited

Loading