For review: YOLO component

[john-]

Per this discussion, I have done a first pass at a YOLO component for Viseron.

Very little testing has been done with the component.. The idea is to get some feedback and if it is worth pursuing make it more robust.

Things to note:

• The ultralytics yolo library is used for loading the model and prediction. There was some investigation into using pytorch directly. However, the ultralytics library is much easier to use for YOLO models and this meets my needs. I am open to revisiting this.
• The user must takes steps to download and install the models themselves. I have not looked into ways of automating this. There may already be a pattern for this in Viseron.
• I have ran the gen_docs script but it does not appear to have filled in everything correctly. For example, _meta.tsx tags array. The docs indicate it should be populated with the tags from the component if they are defined in the component. I do not know where that is defined.

The docs have been updated to reflect how to use pre-trained models.

Link to branch

[john-]

Update: I have now addressed this by creating a new branch and deleting the old one. The link above has been changed to reflect this.

Original post:

I attempted to get updates to the dev branch into my fork and seemed to have messed things up with the git "sync". It looks good on my local machine. Still trying to confirm things are wrong and will make changes from there.

[roflcoopter]

Thank you! Will take a look sometime during the week

• The ultralytics yolo library is used for loading the model and prediction. There was some investigation into using pytorch directly. However, the ultralytics library is much easier to use for YOLO models and this meets my needs. I am open to revisiting this.

Its totally fine to use Ultralytics directly.

• The user must takes steps to download and install the models themselves. I have not looked into ways of automating this. There may already be a pattern for this in Viseron.

Might be better to provide instructions where to download the models. Right now the darknet/edgetpu models are baked into the container, but that creates a lot of bloat for users who dont use those detectors. Some way to download them on-demand would be ideal but im not sure how that would work without hosting the models myself.

• I have ran the gen_docs script but it does not appear to have filled in everything correctly. For example, _meta.tsx tags array. The docs indicate it should be populated with the tags from the component if they are defined in the component. I do not know where that is defined.

When you run it the first time it should detect what domains that you setup. It wont change the file if it already exists tho. You can manually add tags to it, object_detector is the only value i think you'll need

[john-]

I will update the instructions with more detail on model download and installation. Thanks for the info on gen_docs. I have not been able to get the dev container to build without error (Ubuntu and Fedora) and that seems like the better way of running that script so will take another pass at that.

[roflcoopter]

Just managed to reproduce the error, hopefully i can find a solution to the mount failing

[john-]

FYI: It seems like changing the config permissions didn't get things working. I just didn't get an error when I went into vs code after the container trying to load the first time. It turns out that even without changing the permission there is no error on load the second time.

I don't see anything that look like Viseron tasks,

[roflcoopter]

I know whats up now, when using cloneInVolume (which is done when using the instructions on the dev docs) the repo doesnt exist on the host which is why the bind mount fails.

I have cloned it manually which is why it works for me, and my user matches whats inside the container.

Have some ideas how to get rid of the bind mount while still making it simple to use, will get back to you when i have something ready

[roflcoopter]

I think this is fixed now.
I made some changes so instead of bind mounting the config dir dynamically changes in the devcontainer, see the new log statements from the startup here:

************* Setting Viseron config dir *****************
Viseron config dir: /workspaces/viseron/config
*********************** Done *****************************
****** Checking for hardware acceleration platforms ******
OpenCL cannot be used
VA-API cannot be used
CUDA cannot be used
PostgreSQL major version: 14
PostgreSQL bin: /usr/lib/postgresql/14/bin
PostgreSQL data dir: /workspaces/viseron/config/postgresql
*********************** Done *****************************

This required changes to the regular Docker container, so you need to be careful when rebuilding to make sure that the new one is pulled.

I had to stop all devcontainers, remove them along with all images and volumes before i could clone again using the Developer docs.

Edit: Make sure you push your changes first tho as they could potentially get lost when deleting the volumes

[john-]

I think it is working now. I do not get that same error.

This looks pretty benign (cannot remove /usr/lib/locale/locale-archive):

added 834 packages, and audited 835 packages in 59s

237 packages are looking for funding
  run `npm fund` for details

14 vulnerabilities (1 low, 6 moderate, 4 high, 3 critical)

To address all issues, run:
  npm audit fix

Run `npm audit` for details.
pre-commit installed at .git/hooks/pre-commit
Creating default config
Creating frontend .env.local
rm: cannot remove '/usr/lib/locale/locale-archive': Permission denied
Generating locales (this might take a while)...
  en_US.UTF-8...cannot open locale archive "/usr/lib/locale/locale-archive": Permission denied
 done
Generation complete.
Done. Press any key to close the terminal.

After hitting a key I get No such container error:

[484047 ms] Start: Run in container: mkdir -p '/vscode/vscode-server/extensionsCache' && cd '/home/abc/.vscode-server/extensionsCache' && cp 'ckolkman.vscode-postgres-1.4.3' 'eamodio.gitlens-17.1.1' 'esbenp.prettier-vscode-11.0.0' 'github.copilot-1.335.0' 'github.copilot-chat-0.27.3' 'github.vscode-github-actions-0.27.2' 'github.vscode-pull-request-github-0.110.0' 'ms-azure-devops.azure-pipelines-1.249.0' 'ms-azuretools.vscode-containers-2.0.3' 'ms-python.debugpy-2025.8.0-linux-x64' 'ms-python.mypy-type-checker-2025.2.0' 'ms-python.python-2025.6.1-linux-x64' 'ms-python.vscode-pylance-2025.6.1' '/vscode/vscode-server/extensionsCache'
[484680 ms] 
[484681 ms] 
[484681 ms] Start: Run in container: cd '/vscode/vscode-server/extensionsCache' && ls -t | tail -n +500 | xargs rm -f
[484700 ms] 
[484700 ms] 
cannot open locale archive "/usr/lib/locale/locale-archive": Permission denied
 done
Generation complete.
[487225 ms] Start: Run in container: # Test for /home/abc/.gitconfig and git
[487280 ms] 
[487281 ms] 
[487282 ms] Start: Run in container: # Copy /home/johna/.gitconfig to /home/abc/.gitconfig
[487303 ms] 
[487303 ms] 
[487305 ms] Start: Run in container: # Cleaning up git config
[487351 ms] 
[487351 ms] 
[487352 ms] Start: Run: git config --global --get gpg.ssh.allowedSignersFile
[487379 ms] Start: Run in container: command -v git >/dev/null 2>&1 && git config --global --replace-all credential.helper '!f() { /home/abc/.vscode-server/bin/258e40fedc6cb8edf399a463ce3a9d32e7e1f6f3/node /tmp/vscode-remote-containers-bceee1a1-9ede-4548-8bc8-64d7c27d676a.js git-credential-helper $*; }; f' || true
[487403 ms] 
[487403 ms] 
[487405 ms] Start: Run: docker rm -f 0b71cb56d651b8c48fc968640e54f124cda8eca693e2f9359debd40c36e99e00
[487585 ms] Container server terminated (code: 137, signal: null).
[487586 ms] Container server terminated. Reconnecting in 5 seconds...
[487923 ms] Port forwarding 43686 > 42187 > 42187: Local close
[487928 ms] Port forwarding 43668 > 42187 > 42187: Local close
[487928 ms] Port forwarding 43674 > 42187 > 42187: Local close
[487929 ms] Port forwarding 43688 > 42187 > 42187: Local close
[492587 ms] Start: Container: Reconnecting Dev Container server
[492589 ms] Start: Run in container: /bin/sh
[492615 ms] Container server: Error response from daemon: No such container: 0b71cb56d651b8c48fc968640e54f124cda8eca693e2f9359debd40c36e99e00
[492617 ms] Container server terminated (code: 1, signal: null).
[492617 ms] Container server terminated early. Not reconnecting.
[492617 ms] Reconnecting Dev Container server failed: Container server terminated (code: 1, signal: null).

I have not used dev containers before so not sure what to expect.

Instructions say this:

In the future, if you want to get back to your development environment: open Visual Studio Code, click on the Remote Explorer button in the sidebar, select Containers at the top of the sidebar.

This is what I see:

I assume I should look at Dev Containers.

There is no error when I open azure-pipelines in current window.

This works (and everything passes on the current dev branch):

pre-commit run --all-files

So based on this it looks like things are working.

Next, I was going to switchover to my branches to do the pre-commit tasks.

Thanks!

[john-]

Documentation has been updated to include more detail on locating and downloading YOLO models.

[roflcoopter]

Looked through the code and i think it looks good!
What you could do is return False from the setup function in the object detector in case of any failure in setup. That way users will get a better error message incase something goes wrong

[john-]

Thanks for taking a look at it. I will take a look at that return value. After that I will put together a pull request.

[john-]

So I need a bit more direction as to how to do the return of False in case of error. I looked at the other object detectors and they seem to all return True.

Would it be a try/except wrapped around ObjectDetector(vis, config, identifier) ? Sorry if this is a basic question.

[john-]

Thanks for all your assistance with this. The code is now in dev.