UI becomes super slow with > 1K annotations

[H-Shafiei]

Hi

I have a job containing 2 images which I start to annotate with the polygon tool. After annotating more than 800 polygons per image, the UI gets extremely slow. How can this be fixed?

Thanks

[nmanovic]

@H-Shafiei , which version of CVAT do you use? Could you please fill issue template?

[WiktorAdamczyk1]

Hey, I'm having the same issue while annotating on cvat.org. I'm reaching 5000 polygon annotations over 3 images with resolutions around 10000x5000.

My actions before raising this issue

• Searched past issues
• Tried using different browsers (Brave, Chrome, Edge)

Expected Behaviour

UI should be able to handle a large number of annotations on a single image without making every action delayed.

Current Behaviour

Loading an image with around 3000 polygon annotations causes multiple "Page unresponsive" prompts, before said image loads. After an image is loaded every action (hovering over UI elements, padding or zooming the image, editing or creating a new annotation) is delayed by 2-5 seconds.

Possible Solution

After detecting a drop in responsiveness (long image load time, a large number of annotations, or the time it takes to perform certain actions) present a user with a way of increasing responsiveness by allowing the user to choose a smaller part of an image that he currently wants to work on which could work as follows:

• Load the chosen image side by side with the current segmentation mask (prevent "page unresponsive" prompt by not loading annotations) so that the user may choose an area that he wishes to work on.
• Allow the user to choose an area on the image.
• Generate a cropped image and load only the annotations that are in the chosen area.
• Allow the user to annotate the cropped image as normal.
• When saving apply the changes to the original image.

Context

I'm trying to annotate stars on photos made by astronomic telescopes that are usually large files with thousands of visible stars.

Your Environment

I'm using the cvat.org with Brave browser. GPU: AMD Radeon 5700 XT CPU: AMD Ryzen 3600 RAM: 16GB

I assumed that hosting CVAT locally would not provide any improvement however, let me know if I'm wrong.

[pharrellyhy]

Hi, any update on this issue? CVAT becomes super slow when selecting the label if I have few thousands of label candidates.

[Errin890]

Hello, I am also wondering if there is any update on this issue? I regularly deal with images that require thousands of labels and after about 700-800 there is a noticeable lag between every single action (selection, creation, deletion, resizing, moving, etc.), sometimes as a large as whole seconds.

As of this moment, I use the latest version in the develop branch.

[bsekachev]

TODO:

• [ ] Optimize mapStateToProps callbacks, avoid any iteration there [P1]
• [ ] Virtual sidebar with objects [P2]
• [ ] On canvas optimize getBBox() calls if possible, they take significant time. There is also idea do not render elements out of view area [P3]

[josiahls]

Similar related issue:

• Trying to upload coco2017 as a single task
• UI loading is very slow. I'm worried about the speed when I have datasets with 1+ million images and annotations.

Looks like the way around this is to take a dataset, and instead of uploading it whole to cvat, write custom code to chunk the dataset into separate tasks and organize it as its own project

Edit:

• Ended up writing a script that takes COCO 2017 and chunks it into tasks of size 4000 frames. Any higher and the ui loads too slowly when opening the task.

[ThomasMatono]

[ThomasMatono]

What could be the problem

[bsekachev]

@ThomasMatono

That is not related to this issue. Try to look at CVAT logs docker logs cvat_server if you are the instance owner, or contact your system administrator.