Clickable labels instead of typing; others

[gitclem]

I've been playing around with this code (and trying to learn riot js).

Much of the UI mechanics of the program is exactly what I want.

But there are a number of issues that I would like have addressed before it can be used for my company's purposes.

I do recognize this is volunteer effort. I think I can get my boss to help fund this project to get what we want in this program, which is to make it more industrial strength. That is, a high volume (1000's) of labeling and review/fixing that needs to be done.

1. display a choice of labels or feature point names from a predetermined set for the current selected bounding box OR selected feature point. Currently, the list of bounding boxes or feature points is shown AND you have to type the label or feature point name. Typing is time consuming and error prone. Being able to click on the label you want is faster and less error prone. (Also, to output the feature point name instead of a number. I have fixed that in my own version.) Of course, we need a way of pre-defining the labels and later modifying those labels.
2. an easy way to import/export the data into whatever format is being used. The issues are, we will want to change the different types of objects we can recognize. We may also want to remove/add feature points names as we change the model and/or change what we want to recognize. (I don't fully understand how Nimn or the project file works...)
3. store the images within the tree structure of the web app, so that relative path names can be managed and output to the XML file. This is because dlib XML needs to have the relative path names. (If you're managing thousands of image files, putting them all in one directory is untenable.) . The files need to be in the web server's directory structure because of limitations of web servers (due to security). This issue isn't that important to me, as I wrote a python program to patch the imglab generated XML file to fix the relative path names.
4. minor feature: being able to color-code the points by type. This helps the labeler see potential errors. Right now, all the feature points are red dots.
5. minor feature: show list of points (the way it is now) and modify the label or feature name via a popup menu. (I'm trying implement that now, but my ignorance of riot js is slowing me down...)

Watch for changes, or Bookmark for easy discovery.
Fund this project for new features and maintenance.
[Showcase] your project with us by raising an issue

---

Want to back this issue? Post a bounty on it! We accept bounties via Bountysource.

[amitguptagwl]

Thanks for finding this tool suitable for your project. What can it make it more effective is to highlight the users. Currently, approx 4L images gets annotated every month but there is no list of users.

I'm looking for more contributors in term of effort than money because I'm quite busy to make it effective and implement my all ideas about this project.

display a choice of labels or feature point names from a predetermined set for the current selected bounding box OR selected feature point. Currently, the list of bounding boxes or feature points is shown AND you have to type the label or feature point name. Typing is time consuming and error prone. Being able to click on the label you want is faster and less error prone. (Also, to output the feature point name instead of a number. I have fixed that in my own version.) Of course, we need a way of pre-defining the labels and later modifying those labels.

Yes, that is the plan. You may find some issue was already created to get it done.

an easy way to import/export the data into whatever format is being used. The issues are, we will want to change the different types of objects we can recognize. We may also want to remove/add feature points names as we change the model and/or change what we want to recognize. (I don't fully understand how Nimn or the project file works...)

Different formats has different restrictions. Hence, we need a format which can consist all the data so that you can easily use to export in any format. I call it project file. Nimn is the data format discovered by me. So that the project file size is very small and can shared through emails easily. Moreover, the thought was to use it to make multi user collaboration fast. However, due to the business I'm not able to complete that feature completely. Saving the project file in JSON can be more effective till that time.

minor feature: being able to color-code the points by type. This helps the labeler see potential errors. Right now, all the feature points are red dots.
Yes, it can be considered.

minor feature: show list of points (the way it is now) and modify the label or feature name via a popup menu. (I'm trying implement that now, but my ignorance of riot js is slowing me down...)
Riot js is quite easy. Hence, I opted that for this project. I hope you should not take more than a day or 2 to learn it.