As you seem to answer fast ^^, I've some minor improvements that can make user gain some time.

The first time I tried your application, I wrote my first label then I pressed "Enter", which does nothing I know :D
I was hoping it would add the next label input automatically without pressing the "+" button.

I see two alternatives :

• When the user press enter after writing its label, it adds another label, and autofocus on this new label.
• And/Or when the user press the "+" button to add a new label, then it automatically focus on the label text. It will make the user gain one click and some time.

I know it is a really small feature, but I feel it can be useful !
I hope I was clear enough.

Arthur CG

I want to run this tool locally in my machine without the internet, I don't want to upload my data to the server o any, it is exactly like
a labeling tool? but added features? I want to write in python this tool, source code is opensource?

Hi there! I'm updating the official YOLOv5 tutorials, and I wanted to guide people to the best open source labelling tools. I currently only mention Labelbox and CVAT, but I thought I might add this site. I wanted to ask you if any other tools come to mind that you know that would be suitable for YOLO labels?

Secondly, I saw you mentioned you use SSD for labelling, and I thought I'd reach out to encourage you to try YOLOv5. Our smallest and fastest model is 8MB, and reaches [email protected]:0.95, and our most accurate model is YOLOv5x at [email protected]:0.95, which is topping kaggle leaderboards when used for pseudo-labelling (which is what I think you are doing).

If you have any questions, let me know!
https://github.com/ultralytics/yolov5

The experience of the project is great. For improvements, I am thinking it will be good to have a tour prompt that displays next to the important actions, especially as it concerns the following:

• Adding Labels: A prompt next to the + button and disappears once user clicks on the button
• Project Title: A prompt on setting the project title, disappears once user click to edit
• Export: A prompt showing the user the export button after they are done annotations.

See the link below to an example prompt tool from bootstrap.

https://bootstraptour.com/

It would be great when we can add new labels after creating a new project or that we can rename existing labels during the labeling process. This is not an issue, more a feature request.

Just asking whether it will support COCO format in the near future? As Detectron2 is supporting COCO format.

Describe the bug
Failed to compile. I am trying to install and run brand new setup without any changes

To Reproduce
I am trying brand new install without any changes.

Expected behavior
compile
Screenshots
Find the attached screenshot

Desktop (please complete the following information):

• OS: MAC 11
• Browser chrome

Is your feature request related to a problem? Please describe.
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]

I have locally stored folders with images I would like to label. In order to do so, I need to open the folder from the dialog, mark all images. It would be great if I could choose local folders.

Describe the solution you'd like
A clear and concise description of what you want to happen.

Provide functionality to also open local folders.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

Hello,

Thank you for your great tool.
I used it to annotate my images for object detection (rectangles) and then trained a YOLOv3 model using the Darknet repo.

Now, is it possible to add my own model, idk by modifying some tensorflow.js (uploading my model to tfhub ?) code in my fork of this repository and use my own model to annotate other images more easily?

thanks

In my team we are searching for some good labeling stuff and we finally found make-sense. It is very nice looking, easy to use tool, but I have some problems with it.

1. Is there any solution for saving project progress or (better) loading file with image names and labels just to keep labeling after restart of computer?

2. Is there (or will be) any solution for sharing work with many people labeling at the same time?

3. It will be nice to have information how many photos was labeled and how many are still to do.

Thank you for any answer.

Is your feature request related to a problem? Please describe.
I want to annotate a small set of pictures with "oriented bounding boxes". I have been using this tool (BBox with angle label) and it works fine, but the interface is counter-intuitive (draw the box, then rotate). As a result I looked into options for a better interface, but did not find anything.

Describe the solution you'd like
The aforementioned tool outputs six scalars. It would be nice to have similar output but with a better interface (i.e. the make-sense UI).

    BBox_num :number of bounding box (there may be multiple per picture)
    xc: rectangle center x
    yc: rectangle center y
    w: width of the rectangle
    h: height of the rectangle
    theta: angle of the rectangle

Describe alternatives you've considered
Well, I can continue using current tool, so it's not too critical, just a would-be-nice if I label hundreds more images. It's most time efficient for me to keep using the current tool, as actually it only takes an hour or two per 100 images, and this is simply a small / preliminary research project so at most ten hours will be actively put into this. But since this is open source, I would be happy to contribute to developing this functionality into makesense in case it's useful to others.

Additional context
For context, interest in oriented bounding box comes from the aerial imagery research field.
https://captain-whu.github.io/DOTA/dataset.html HOWEVER these are a bit different as they allow for any quadrilateral, not just right angle rectangles as I am working with.

Dota is a large-scale dataset for object detection in aerial images.

I am not working in that field, just taking advantage of algorithms such as YOLO-OBB.

When you want to create public open datasets, you want people to collaborate, and join as fast as possible, for this, make-sense is ideal because it's a web project, it's downsides is that you have to make sure that everyone that wants to collaborate has to have the files already downloaded, in which I can consider ineficcient because you can perfectly load up image files from the browser.

Describe the solution you'd like

It'd be great if make-sense allowed you to give a yaml or a list of urls that are available, so that you can colaborate with people on the same set of images

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.
sure, you could theoretically mount your image database as a FUSE point in your filesystem in order for everyone to share the same set of images but you'd have to substantially edit the host filesystem, which is hard, so what I would like is for example you can just link a pastebin/gist.github/github repo/gitlab/your_host_here with a link of all images, these of course is useful for projects where the images are public, which many projects are. These would increase colaboration for open datasets, so that you can jump in into annotation with no prep time at all!

Additional context
Imagine you could get started on Openimages v6 (or more realistically, a subset of it) without you having to download the images yourself, isn't that the point of web-apps

This of course would make loading existing annotations easier since the images are stored on some server on the web.

For keypoint/landmark annotation I think it would be great if the editor area would keep the keypoints from the last image when moving to the next one. As most of the times you want the same number of keypoints but you have not only recreate all the points but also assign the labels this is very cumbersome when you have many points lets say more than 10. If you already have all the label assigned points you simply need then to move them to the right place this is a lot faster than clicking them again and assign labels especially if you have images that are very similar so the keypoint might only move by less than 20 pixels between images.

The tool is great, but the object detection utility is an overkill for image classification.
Are you gonna add the possibility to use make-sense for image classification.

Thanks for the great work!

Describe the bug
A clear and concise description of what the bug is.
I load an image, make points annotation, then export the csv file. There is nothing inside the csv file.
I then draw a bounding box on the same image, then export the points. This time the csv file contains the points locations.

To Reproduce
Steps to reproduce the behavior:

1. Load an image
2. make some points annotations
3. export the points to a csv file

Expected behavior
The csv file should contains points locations even if there is no bounding box.

Screenshots
If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• OS: Ubuntu 16
• Browser: Chrome
• Version: 70

I'm the user of another opensource image labeling application https://github.com/NaturalIntelligence/imglab.

I noticed that the number of stars to your application are increasing every day quickly. But I found that this application is the kind of subpart of https://imglab.in (feature-wise). So am I missing something? or do we need to put the effort to create another image labeling tool?

Hi, I am trying to use make-sense for my own object detection project. Yet, when I want to upload the csv to Google Cloud. It seems like not really supporting. Just wondering if a new feature can add to export format fits GCP/ AWS? Thanks

Hi,

I am annotating a new image using Polygon Annotations. But when I am downloading the images with Yolo format it's not downloading the .txt file for the image in which I have annotated with polygon boxes.

Thank you

npm start

[email protected] start /Users/makamkirankumar/Downloads/make-sense
react-scripts start

Attempting to bind to HOST environment variable: x86_64-apple-darwin13.4.0
If this was unintentional, check that you haven't mistakenly set it in your shell.
Learn more here: https://bit.ly/CRA-advanced-config

dyld: lazy symbol binding failed: Symbol not found: _FSEventStreamCreate
Referenced from: /Users/makamkirankumar/Downloads/make-sense/node_modules/chokidar/node_modules/fsevents/lib/binding/Release/node-v79-darwin-x64/fse.node
Expected in: flat namespace

dyld: Symbol not found: _FSEventStreamCreate
Referenced from: /Users/makamkirankumar/Downloads/make-sense/node_modules/chokidar/node_modules/fsevents/lib/binding/Release/node-v79-darwin-x64/fse.node
Expected in: flat namespace

npm ERR! code ELIFECYCLE
npm ERR! errno 1
npm ERR! [email protected] start: react-scripts start
npm ERR! Exit status 1
npm ERR!
npm ERR! Failed at the [email protected] start script.
npm ERR! This is probably not a problem with npm. There is likely additional logging output above.

npm ERR! A complete log of this run can be found in:
npm ERR! /Users/makamkirankumar/.npm/_logs/2020-02-24T17_12_19_308Z-debug.log
(base) Makams-MacBook-Pro:make-sense makamkirankumar$ events.js:298
throw er; // Unhandled 'error' event
^

Error: write EPIPE
at process.target._send (internal/child_process.js:811:20)
at process.target.send (internal/child_process.js:682:19)
at RpcProvider._dispatch (/Users/makamkirankumar/Downloads/make-sense/node_modules/fork-ts-checker-webpack-plugin/lib/service.js:20:17)
at /Users/makamkirankumar/Downloads/make-sense/node_modules/worker-rpc/lib/RpcProvider.js:107:52
at processTicksAndRejections (internal/process/task_queues.js:97:5)
Emitted 'error' event on process instance at:
at internal/child_process.js:815:39
at processTicksAndRejections (internal/process/task_queues.js:79:11) {
errno: -32,
code: 'EPIPE',
syscall: 'write'
}

Expected behaviour

When drawing a bounding box or other marker over the edge of an image, anything drawn over the edge clips to the edge.

Observed behaviour

When drawing a bounding box over the edge of an image, the drawing cancels. This makes it more difficult to label objects in the periphery of images, which is frequently where they will be in microscopy images.

Hey, first off, thank you for putting in the time to create this great tool. I do have a very small feature request - I think adding tool-tips for the top action-bar buttons would help with usability (especially with the accept/reject all buttons).

Thank you very much for this amazing tool.

When I try to label images with a white background, the current white guidelines make me feel difficult to locate the cursor. Is it possible to add an option to change the guideline colour?

Thank you.

I have 8 different labels.
When I add more than 20 (more or less, it depends on the screen) boxes on one image. If I add one more bounding box. I won't be able to scroll down to all labels. I'll be able to pick only first ones.

See the screenshot linked. I have more labels under wheat but I can't access them.

**Desktop **

• OS: Windows 10
• Browser : Firefox
• Version : 60.0.2 (32 bits)

Is your feature request related to a problem? Please describe.
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]

I would like to load previously defined Labels (which would be technically possible). Unfortunately I don't know how this input has to look like/be formatted.

Describe the solution you'd like
A clear and concise description of what you want to happen.

Please provide an example file.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

First, this is a very remarkable project and I am looking forward to it making significant improvement in the image annotation experience.

It will be great if support for exporting annotated data to Pascal VOC XML can be provided. See the links below for example blueprint of the format.

https://gist.github.com/Prasad9/30900b0ef1375cc7385f4d85135fdb44

https://github.com/AndrewCarterUK/pascal-voc-writer/blob/master/pascal_voc_writer/templates/annotation.xml

There is an existing non-UI tool that does that. You can take insights from it to implement the UI based annotation. See the link below for the tool.

https://github.com/AndrewCarterUK/pascal-voc-writer/blob/master/pascal_voc_writer/templates/annotation.xml

Is your feature request related to a problem? Please describe.

Before the process of annotation, I believe many people go through the process of data cleaning: things like image-level labels (brightness, image quality,...). Furthermore, there might be a need to crop the images with a certain aspect ratio with a certain scale for the image.

Describe the solution you'd like
Having a labeling and cropping tool for data cleaning with a batch ofdata.

Is your feature request related to a problem? Please describe.
re-edit existing annotations

Describe the solution you'd like
add an import button to load existing annotations

Describe alternatives you've considered
No

Additional context
No

Would it be possible to drop some noisy images from the dataset?

Is your feature request related to a problem? Please describe.
Pressing 'back' button (or backspace key on some browsers) causes the browser to navigate back, as expected, which in turn causes all labelling work to be lost and unrecoverable.

Describe the solution you'd like
Prompt the user whether the really want to go 'back' (and lose their work) or halt. e.g. https://stackoverflow.com/questions/12381563/how-to-stop-browser-back-button-using-javascript

Describe alternatives you've considered
Alternatives are to save the work in some way, but given discussions about saving/reloading/persistent labelling, this seems harder to do than preventing back button.

Additional context
Tested using Firefox v79 on Mac

Is your feature request related to a problem? Please describe.
I can't save my work. My dataset contains more them 1K images - it is impossible to annotate them in a day therefore I'd like to save my work and then return to one the next day - but I can't find something like Save or project, or something which allows me to return to my work which is already done

Describe the solution you'd like
Any mechanism which allows me to save my work

Describe the bug
While using your app I have noticed the following visual bug(visible in the attached screenshot).
When the browser's download bar is visible and listing downloaded files, the image is partially covering the project's title and two texts are overlapping.
I hope it was helpful for the project

All the best, keep it going, never give up bye :)

Screenshots

in image recognition task,can a image supporting by multi labels,not just one label.

Is your feature request related to a problem? Please describe.
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]
If your model is open, and you want to use them on the web, to get automated suggestions, make-sense should support it.

Describe the solution you'd like
A clear and concise description of what you want to happen.
When opening the app, there should be an option to put your own model.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.
An alternative would be preprocessing your images with your private model and then load the annotations to make-sense but as far as I know, it's not supported either.

Is your feature request related to a problem? Please describe.
Problem: All progress is cleared when reloading the page, which equals bad user experience.

1. I have been trying to implement persistent state with redux-persist but I have run into this problem:
   "TypeError: Cannot read property 'width' of null"

Function.calculateImageScale
src/utils/RenderEngineUtil.ts:12
   9 | 
  10 | export class RenderEngineUtil {
  11 |     public static calculateImageScale(data: EditorData): number {
> 12 |         return data.realImageSize.width / data.viewPortContentImageRect.width;
  13 |     }
  14 | 
  15 |     public static isMouseOverImage(data: EditorData): boolean {

2. While getting comfortable with the codebase it seems there might work to be done to reload the images from the persistent state e.g. reloading the page

Describe the solution you'd like
I would like for the progress to be persistent and to not be cleared when reloading the page.

Describe alternatives you've considered
I have also tried localStorage and the same issues from above occurred.

Additional context
Check out the branch redux-persist from my fork https://github.com/JorgeCeja/make-sense

This would be part of an automated computer vision project

-add option to zoom in and out of image
-this will help in creating more accurate annotation

Describe the bug
I loaded two images in the project, make annotations on both images, then export the csv file. There is no file name in the csv data, so it is impossible to tell which points locations belongs to which image

To Reproduce

1. Load two images
2. Make points annotations
3. Export as csv file

Expected behavior
There should be a filename column in the csv file. Like this:
file1.jpg,lane1,217,556,01080.jpg,1640,590
file1.jpg,lane1,314,503,01080.jpg,1640,590
file1.jpg,lane1,410,458,01080.jpg,1640,590
file1.jpg,lane1,492,415,01080.jpg,1640,590
file1.jpg,lane1,564,383,01080.jpg,1640,590
file1.jpg,lane1,633,349,01080.jpg,1640,590
file1.jpg,lane2,1581,454,01080.jpg,1640,590
file1.jpg,lane2,1486,430,01080.jpg,1640,590
file1.jpg,lane2,1402,411,01080.jpg,1640,590
file1.jpg,lane2,1263,378,01080.jpg,1640,590
file1.jpg,lane2,1171,353,01080.jpg,1640,590
file2.jpg,lane1,217,556,01080.jpg,1640,590
file2.jpg,lane1,314,503,01080.jpg,1640,590
file2.jpg,lane1,410,458,01080.jpg,1640,590
file2.jpg,lane1,492,415,01080.jpg,1640,590
file2.jpg,lane1,564,383,01080.jpg,1640,590
file2.jpg,lane1,633,349,01080.jpg,1640,590
file2.jpg,lane2,1581,454,01080.jpg,1640,590
file2.jpg,lane2,1486,430,01080.jpg,1640,590
file2.jpg,lane2,1402,411,01080.jpg,1640,590
file2.jpg,lane2,1263,378,01080.jpg,1640,590
file2.jpg,lane2,1171,353,01080.jpg,1640,590

Screenshots
If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• OS: Ubnuntu 16
• Browser: Chrome
• Version: 70

Thanks for the great tool; well-done, and I will be looking forward to trying out new features.

One note for Windows 10 Users. After following the instructions and successfully installing the make-sense I couldn't view it in the browser (e.g. Google Chrome). Basically npm start in Windows 10 Ubuntu results in the following error in the browser:

Error: Cannot find module '@csstools/normalize.css' nodejs

Although the message is self-explanatory, but I was struggling a bit since I have had zero experience with Node.js. Finally I figured one need to install normalize.css simply:
npm install normalize.css

Maybe it is best that README to be updated.
And you are good to go, Happy labelling!!

It would be great to support a faster annotation scheme where a bounding box is drawn by two clicks (the first is the top left and second is the bottom right).

Polygon's should implement follow the standard polygon spec. Doing so significantly reduces the complexity of etl on polygons because they will be valid in libraries like shapely

Hi,

You have two Citation part in README.md

Thanks for providing such an annotation tool. In my project, I am trying to annotation line, but I find there isn't the option, would you mind telling me what I should modify or add in the current code base to support this idea?

I find labelling a very time consuming task, that's while a usual approach is to onboard more people to create both bigger datasets and consensus on labelled objects.

A way we could unite this with the current development of the tool is via creating a sort of "task" distribution, where a subset of the data can be added to a session and this sent to a specific labeler. That would free the need to have a user system (keep it lean and simple) and still distribute the tasks around. Could be potentially be set up through an API (but that's another feature request).

Is your feature request related to a problem? Please describe.
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]

Describe the solution you'd like
A clear and concise description of what you want to happen.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

Describe the bug
The tool is very slow when we zoom in and make bounding box.

To Reproduce
Steps to reproduce the behavior:

1. upload images
2. zoom max in image
3. make bounding box

Expected behavior

Screenshots

Desktop (please complete the following information):

• OS: Windows 10 pro
• Browser modzilla firfox
• Version 69.0 (64 bits)

Smartphone (please complete the following information):

Additional context

It'd be useful to see some running statistics like the number of labeled objects in each class and the number of images labeled so far.

When i'm labeling i always want to have these numbers in mind.

I found that labelbox (another labeling online tool) has this kind of statistics:

could you think of something like this?

Thank you for your good work

Hey :)

For my personnal use I face the problem that I have to parse again the export file to modify it to have a format I use on my AIs.

I would like to choose on exporting file, the things I want to export and how I want it.
For example, I'm working on a project where my labels need to be on this format :
Filename.jpg x_min, y_min, x_max, y_max, class x_min2, y_min2, x_max2, y_max2, class2 etc... (if I have multiple anchors on my .jpg)

I can imagine on the export pop-up a new tab with the format I want to define. Filename, x coordinates, y coordinates, width, heigth, etc ...

Of course I can develop a tool to convert it how I want, but I thought it could be usefull to add it.

Btw, great job. Your app rocks !

is this make sense AI a fully self hosted solution i.e. it can all be run locally?

Thanks

It is great to see someone picking up the visual data annotation problem seriously.
However, you are obviously not the first one who has been trying that. I've tried and checked quiet a few data annotation tools in the past and I noticed they are all very weak if at best barely usable when it comes to doing segmentation masks.

As either they go one of the two potential routes:

*Only allow polygonal input to mask polygons

*Only allow painting input via a pixel brush

But both of them have issues. Paint brushes are cumbersome if you wan to mask large areas and take a lot of time.
Polygons have an issue if you want to mark small and thin image regions as you simply would have to define so many points and often you can not just hit the edge correctly.

Best would be able to mix both of them so you can draw polygons and draw above the polygons in a multi layer fashion. Or if that is too complicated allow the option to bake the polygons down into the segmentation mask image and then have the option to use a paint brush to paint on top.