hitachi-automotive-and-industry-lab / semantic-segmentation-editor Goto Github PK

Hi, I use your editor for labeling pcd point cloud data.
I made point cloud data file using drone and pix4d program.
There are x,y,z and r,g,b and scalar information in my point cloud data.
what data is in your sample(pointcloud_labeling.pcd)?

I am looking for an segmentation tool which will filter only the point cloud that I am interested in. Take the demo point cloud sever as an example. I have segmented out the car manually. But when I click the "PCD output", I get a PCD file with 0kb size. Is this a bug, could I output a PCD file with only the point cloud I am interested in?

hi Damien Mandrioli:
i want to use this tools to building my own pcl dataset , on VM(ubuntu16.04),and windows 10 install ,all fail,but the docker can run . i wanna change the calss name in label pcl ,how can do that!

thanks for your help!

Thank you for providing a good tool.

At the end of the labeling process, i press the 'J' key to view the json file.

I can see the json file, but i want to download json files.

How do I download automatically?

hi，The video can not be loaded in vimeo，is there any other choice to play the video？
thank you！

I have successfully installed Meteor using Chocolatey but I am unable to run the tool. Could you please suggest how to proceed further using this tool?
I am using Windows OS.

Is there a function that support json that descript polygon to png annotation image used for segmentation training?

hi, guys.
very good work for semantic segmentation labelling.

Is there a fast way to load next example when finish labelling the current one?
Currently, I have to go back to the file browser, then click some image to load next one, which reduce the label speed.

A fast hotkey is needed for save current example and load next one.

Currently only plain text PCD files are supported

I'm trying to annotate an xyzrgb pcd file (some plants) via this tool.But I found that it is difficult to distinguish where the background and where the leaves is.If the point has its origional color,it would be more easier.Maybe a botten to switch the view between colored view and labled view would help?

Thanks for providing such a great tool for point cloud data edit, but i want to konw is there any possible to change some code that i can get the information of color after processing?

When I try to run the app with "meteor npm start" I get the following:

`Number of available sets of object classes: 3
=> Started your app.

=> App running at: http://localhost:3000/
`

but I can`t see any of the editor windows.

Does anyone know how to fix this?

I`m using semantic-segmentation-editor-1.1.2 on Ubuntu 16.04.

You have down a good job. But when I want to use it to train SSD, I cann't find a way to export VOC-like xml format dataset. The only support export format is json.

When drawing polygon on my images, "Enter" button does not work and "Esc" key does not work too.

what is the result of the 3D point cloud editor? How could I know the size of 3D bounding box? is that in the xxx.pcd.labels?

The error is :
Error: ENOENT: no such file or directory, stat '/home/wenjingw/pcd_annotation/semantic-segmentation-editor/output/1532169145283189.pcd.objects'

and I cant load pcd file.

Thank you for good tool.
I would like to load auto-annotated json files.
Is it possible?

I might have missed it, but is there a way to export just the segmentations as images? Downloading the image saves the image with both the original and overlaid segmentations. I'm aware of the JSON format, but is there an easy tool for translating that into a segmentation image?

Thanks for sharing this great tool. I actually run into some problems when running the docker image. Can't find the Dockerfile in the repo and on dockerhub. Could you share it? Thanks!

How to save 2D images Segmentation results ?
I don't know what mean exactly ---- (Just use API)
Should I save json file by hand?

i tried drawing Rectangle ,then i tried selecting drawn rectangle its selecting but if i trying edit/drag only one corner is dragging

Hi Team,
This is really an amazing tool. Good job.
I am facing one issue. If I select the points for a class (say Car) these are selected in red color. I can remove a particular point which does not belongs to this Car class by using (-) button provided in the tool. All works perfect till here.
Now, when I finally add these points to the car class, these points color changes to blue. Now, If I want to remove any point (which was marked mistakenly) from this class, I cannot do that.
Could you please help with this?

Hi,
Thank you for providing such a good tool.
When I output labeled pcd file from a unlabeled pcb including intensity, the output labeled data doesn't include intensity and only XYZ. Is it possible to keep XYZI when output pcd file?
I don't know how to do it, could you please help with this?

maybe a little stupid but I tried for a while and stilll don't know where to use API command...Is that truly a command ? Under which directory should I input that command ?

How to export the labels ? sorry for I didn't find where to export all the label json.
And the Show Json output button failed to show the json 😢

Hi, thanks for sharing this great tool. Currently the only way to navigate to next/previous image is going back to the browsing mode and select the image. I think it's fine because it only costs a small fraction of time comparing to finish a segmentation label. However, I found it become much slower when I have more than a thousand labels in the MongoDB. Then I checked the requests:

and also the preview of the response

It downloads every single record in the system every time when going back to the file browsing mode. So the whole system will become slower and slower over the time. I'm trying to figure out how to fix this issue, but I'm not familiar with some part of the tech stack used in this project. Would be grateful if someone could help.

** I use the suggested docker image jshimko/meteor-launchpad:latest and chrome browser

Hi, thanks for creating this tool!

I'm running sse in my own browser, and trying to draw a polygon on my image. However, after selecting the polygon tool and clicking on the image, nothing happens. It's the same with rectangle tool and magic tool.

I do not get any error statement in browser or terminal. I tried inserting a console.log statement in SsePolygonTool.onMouseDown, and it seems that the function is not triggered when I click on the image.

It would be really great if you can help diagnose the issue.

Hi, I faced another performance issue. When label number approaching ten thousand I found the speed of entering editing mode becomes much slower again. Then I check the network and see this response.

I grep the keyword "removed" and found nothing meaningful. Could you give me a hint about how to fix this? Thanks.

Hi Team,
It will be great if you add and Undo/Redo option for PointCloud editing tool. Whenever I mistakenly select a large number of point (using right click), I have to manually undo them all. An Undo option will be great to clear my last action performed.
And similarly for Redo.

@dmandrioli i know how to use tha eiitor follow the readme.txt, but i'm confused how to imput my data, i didn't find any information about how to input my own point cloud data. wolud you please tell me how to solve the problem.
Thanks, looking forward your reply!

Hello, i have a black screen when running the app even if the input folder contains .jpg images.
Permissions on the input folder are set to drwxrwxrwx

No error in logs.

macOs 10.13.6
[email protected]

I tried to read the data as:

labels= [int(line) for line in f]
and the feedback was:

File "/usr/lib/python3.5/codecs.py", line 321, in decode
(result, consumed) = self._buffer_decode(data, self.errors, final)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x82 in position 4: invalid start byte

So how can I get the data?

After spenting 2 hours of install meteor and depencies, I succeed to build it. But when try to run, I got this error.
Is there any suggestion?
W20180730-18:21:40.261(8)? (STDERR) Error while parsing config.json
W20180730-18:21:40.348(8)? (STDERR) C:\Users****\AppData\Local.meteor\packages\meteor-tool\1.6.1_3\mt-os.windows.x86_64\dev_bundle\server-lib\node_modules\fibers\future.js:280
W20180730-18:21:40.350(8)? (STDERR) throw(ex);
W20180730-18:21:40.350(8)? (STDERR) ^
W20180730-18:21:40.351(8)? (STDERR)
W20180730-18:21:40.352(8)? (STDERR) TypeError: Cannot destructure property imagesFolder of 'undefined' or 'null'.
W20180730-18:21:40.352(8)? (STDERR) at api.js (server/api.js:13:61)
W20180730-18:21:40.353(8)? (STDERR) at fileEvaluate (packages\modules-runtime.js:343:9)
W20180730-18:21:40.353(8)? (STDERR) at require (packages\modules-runtime.js:238:16)
W20180730-18:21:40.354(8)? (STDERR) at D:\Detection\AnnotationTools\semantic-segmentation-editor.meteor\local\build\programs\server\app\app.js:1015:1
W20180730-18:21:40.354(8)? (STDERR) at D:\Detection\AnnotationTools\semantic-segmentation-editor.meteor\local\build\programs\server\boot.js:411:36
W20180730-18:21:40.357(8)? (STDERR) at Array.forEach ()
W20180730-18:21:40.358(8)? (STDERR) at D:\Detection\AnnotationTools\semantic-segmentation-editor.meteor\local\build\programs\server\boot.js:220:19
W20180730-18:21:40.359(8)? (STDERR) at D:\Detection\AnnotationTools\semantic-segmentation-editor.meteor\local\build\programs\server\boot.js:471:5
W20180730-18:21:40.368(8)? (STDERR) at Function.run (D:\Detection\AnnotationTools\semantic-segmentation-editor.meteor\local\build\programs\server\profile.js:510:12)
W20180730-18:21:40.372(8)? (STDERR) at D:\Detection\AnnotationTools\semantic-segmentation-editor.meteor\local\build\programs\server\boot.js:470:11
=> Exited with code: 1
W20180730-18:21:46.480(8)? (STDERR) Error while parsing config.json

They should be red when pushed.

The edit functions works fine but json api failed

In the output_dir, generate xxxx.pcd.labels and xxxx,pcd.objects , how to use the output data?

Sorry to bother you, but this error drive me crazy, I did as the "readme.txt", and every time I input

meteor --settings settings.json

and blow come out

Your app is crashing. Here's the latest log:
/Users/apple/Documents/NN_Models/semantic-segmentation-editor/.meteor/local/build/programs/server/boot.js:475
}).run();
^
Error: Cannot find module '@babel/runtime/helpers/builtin/interopRequireDefault'
at Function.Module._resolveFilename (module.js:547:15)
at Function.resolve (internal/module.js:18:19)
at Object.require (/Users/apple/Documents/NN_Models/semantic-segmentation-editor/.meteor/local/build/programs/server/boot.js:288:32)
at makeInstallerOptions.fallback (packages/modules-runtime.js:651:18)
at require (packages/modules-runtime.js:244:16)
at livedata_connection.js (/Users/apple/Documents/NN_Models/semantic-segmentation-editor/.meteor/local/build/programs/server/packages/ddp-client.js:147:30)
at fileEvaluate (packages/modules-runtime.js:343:9)
at require (packages/modules-runtime.js:238:16)
at namespace.js (packages/ddp-client/common/namespace.js:1:300)
at fileEvaluate (packages/modules-runtime.js:343:9)
Exited with code: 1
Your application is crashing. Waiting for file change.

I google it , and someone say I should install babel, and I did it, but nothing happened, wait for your reply!

Is there any way that we could get some support for masks - i.e. binary or multiple class segmentation so each pixel is labelled at the category level (e.g. background 0, car 1, bicycle 2 etc).

1. I work 174 images & click next button

2. Next img is displayed as null.
   

3. http://localhost:3000/browse/1/20/null -> http://localhost:3000/browse/1/20

worked.

It is bug?

i want to keep the intensity value of the pointcloud. but the software erase it. How can i keep the intensity property

hello,
in the beginning, I really want to thank you for this great tool. it saved me a lot of time and effort

i got a problem when i press the download button after I segmented an image.

it is shown in this screenshot

Response headers need to be fixed

I've annotated a couple hundred images using the tool and plan to do more. In order to download the annotated image, I click download for each image. Is there a way to just download all images at once?

Are there any plans to merge your annotation tools to support the ability to automatically label a velodyne scan by back projecting the labels in the corresponding image?

Hi, all I've install npm with version 1.6, and it's ok, after command
"meteor npm install
meteor --settings settings.json",
and I sucesfully change the setting.jason, and push some pcd file into the folder, after I log into the web browser, why no pcd visualization.? this is strange.
and it appears in the terminal like:
W20180920-17:23:57.041(8)? (STDERR) Error: ENOENT: no such file or directory, stat '/home/jean/semantic-segmentation-editor/mnt/image2/table_scene_lms400.pcd.labels'
W20180920-17:23:57.137(8)? (STDERR) Error: ENOENT: no such file or directory, stat '/home/jean/semantic-segmentation-editor/mnt/image2/table_scene_lms400.pcd.objects'

OUTPUT_FOLDER is no files after labeled