Very high RAM usage of CVAT in browser for object detection bounding box annotations on client side.

[YadneshD]

Actions before raising this issue

• I searched the existing issues and did not find anything similar.
• I read/searched the docs

Steps to Reproduce

I have deployed CVAT on a server with high RAM config around 32 GB. But on client side, in browser, the RAM available is 6 GB. In a few projects, the RAM on client PC is exhausted by the browser running CVAT. I don't have higher RAM config for annotators working on my object detection project. They are editing bounding boxes on a object detection project. 6 GB is still decent RAM for annotating image frames. Please help! The PC gets hanged and freezes when I load few images from that project.

Expected Behavior

No response

Possible Solution

No response

Context

No response

Environment

Client side - 
Windows
Chrome broswer
6 GB RAM.

Server side -
Linux Ubuntu 22.04
docker

[bsekachev]

CVAT version? And what is typical image resolution in the project?

[YadneshD]

Server version: 2.19.0

Core version: 15.2.0

Canvas version: 2.20.9

UI version: 1.66.0

Few Images are HD of 6 MB size and 5000 x 8000 pixels resolution.

[bsekachev]

And how much RAM does the browser use?

You may know it for example by hovering Google Chrome tab with the mouse

[YadneshD]

Currently the annotators PC have 16 GB RAM. Only chrome is running on it. Sometimes it consumes 15 GB RAM which is much more for simple annotations job. For some images, even 16 GB RAM crashes.

[bsekachev]

I am interested in how much RAM tab with CVAT uses, not Chrome itself. Usually Chrome does not allow more than 4 GB for one tab. But it depends on system configuration.

[YadneshD]

Chrome shows CVAT consumes max 6.7 GB. At start, it is 316 MB, then fluctuates between values - 5.3 GB, 6.7 GB, 3.3 GB, etc. when I navigate between images. The RAM utilization got from 15 GB to 5.1 GB when I closed chrome on the whole.

[bsekachev]

So, you just have a task with ~5000x6000px images (btw what is chunk size? or you did not change its default value?), you open the job and navigate between frames? Do you do anything else? (e.g. apply color filters, image equalization or something else?)

Also, you have contextual images in the task?

May you provide a backup file, containing the task as a minimal reproducing example?

[YadneshD]

I didn't change default value. Yes I need to navigate frames sometimes. Don't apply any color filters, equalization or anything else.
What are contextual images ?
I need some time to have the example. I will provide you the zip file by tomorrow.

[bsekachev]

Okay, I will wait then.

[YadneshD]

@bsekachev Please download sample dataset - download

It is YOLOv8 detection 1.0 format.

[YadneshD]

@bsekachev Any updates on it buddy? Are you guys working on it ? Have you guys downloaded sample dataset from above link provided? I need to delete it from drive.

[bsekachev]

I downloaded the file, but did not check the issue yet.