Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Yolo #178

Open
Charelvanhoof opened this issue Dec 30, 2020 · 4 comments
Open

Yolo #178

Charelvanhoof opened this issue Dec 30, 2020 · 4 comments

Comments

@Charelvanhoof
Copy link

Hi,

Thank you for organizing this for Robocup. I did upload some test images to test the platform and see if it would benefit for us.
Some points I noticed, might be a feature request or maybe something I missed.

  • I would like to export in Yolo format or VOC XML format ( or some tensorflow format for SSD MobilenetV2, etc) and I was looking for the possibility to export the annotations in these more common/standard formats. Do you have these ready available? Also in e.g. yolo format I need a file per image, not one large file.
  • Did some test with annotation of the ball and robot, that worked fine. But when I selected to annotate a goal-post I got a different method. I could not find the option to change it and have goalposts the same way annotated as e.g. ball or robot?
  • We are also considering image segmentation, would that be supported with some advanced functionality to auto-trace objects?
  • Do you have some integration planned with tools like labelimg?

Thanks
MSL team VDL robotsports

@greeniehemi
Copy link

Are you facing an isssue with labeling the different classes of image?

@Charelvanhoof
Copy link
Author

I guess the question relates to "Did some test with annotation of the ball and robot, that worked fine. But when I selected to annotate a goal-post I got a different method. I could not find the option to change it and have goalposts the same way annotated as e.g. ball or robot?"
I did bounding boxes for ball and robot, but for goal-post tags I suddenly got a polygon (I guess) but it is not something I could change?

@timonegk
Copy link
Member

Hi, I hope that I can answer your questions:

  • I would like to export in Yolo format or VOC XML format ( or some tensorflow format for SSD MobilenetV2, etc) and I was looking for the possibility to export the annotations in these more common/standard formats. Do you have these ready available? Also in e.g. yolo format I need a file per image, not one large file.

As far as I know, the formats you mentioned are not yet available. You can try to add them when defining a custom export format; the export format definition is pretty extensive. Generating multiple files is currently not possible, though. Maybe it would be possible to implement this feature directly in the ImageTagger that would probably be useful for a lot of people. I can look into that.

The way we currently use the export format is that we export the data as yaml (there are already export formats for that) and then process the file locally.

  • Did some test with annotation of the ball and robot, that worked fine. But when I selected to annotate a goal-post I got a different method. I could not find the option to change it and have goalposts the same way annotated as e.g. ball or robot?

There are a few different methods of annotating. The method for balls, robots and goals is the bounding box method. Goalposts however use the polygon method. That is a decision that we took for the publicly hosted instance because goal posts often are tilted in the image and we did not want to lose that information (it cannot be represented with bounding boxes).
There are additional method, for example “point” which is used for line intersections and ball centers and the additional methods „line“ and „multiline“.
In a self hosted instance, the methods for different annotation types can be manually defined. In the public instance on https://imagetagger.bit-bots.de we decided for the methods that we or other teams prefered.

  • We are also considering image segmentation, would that be supported with some advanced functionality to auto-trace objects?

Image segmentation is currently not supported and also not a planned feature. But if you would like to add it, feel free to ask for help!
Auto-tracing of objects (for the currently existing methods) is actually something we plan on implementing in the near future.

  • Do you have some integration planned with tools like labelimg?

There is currently no integration with labelimg but I guess improving an import/export flow to and from labelimg would be very nice. I see that labelimg saves the annotations in the PASCAL VOC format. Adding the option to upload this format should be doable. I can look into that sometime in the next weeks.

I hope that this helped!

@Charelvanhoof
Copy link
Author

Thank you for the quick answers!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

3 participants