integrate yolo as ML backend into my label tasks

[JasonSloan]

Hi, i tried to integrate the ML models into my label tasks, everything seemd to work fine but no predictions showed though.

Env: label-studio was running in a docker container while the ML-backend was running in the container of the container, but i checked the ENV variable as well as the mapping directory inside the ML-backend container and they were all correctly set.

Tests: I tested in my web browser of the connection between label-studio and ML-backend and it worked fine.

Model weight: I put an official yolov8m.pt file downloaded from github into the models folder and i checked inside the ML-backend container that it was really being there.

Settings: I set the label template to yolo and wrote the configurations code below:
<View> <Image name="image" value="$image"/> <RectangleLabels name="label" toName="image" model_score_threshold="0.25" model_path="[yolov8n.pt](http://yolov8n.pt/)" opacity="0.1"> <Label value="Person" background="red"/> <Label value="person" background="red"/> <Label value="Car" background="blue"/> </RectangleLabels> </View>

Finally, i checked the images inside the project but no person bounding boxes were shown. Besides, I was pretty sure that the yolov8m.pt can successfully detected them.

So would you please tell me that where i was doing wrong?

[makseq]

Hi,

1. This line looks wrong in your labeling config

 model_path="[yolov8n.pt](http://yolov8n.pt/)"

it should be

model_path="yolov8n.pt"

2. You don't need to download official models manually, they will be downloaded automatically.
3. Most likely, you didn't specify LABEL_STUDIO_URL and LABEL_STUDIO_API_KEY in the docker-compose.yml file.
4. Check ML backend logs, the problem should be mentioned there.

[JasonSloan]

Hi,

1. This line looks wrong in your labeling config

 model_path="[yolov8n.pt](http://yolov8n.pt/)"

it should be

model_path="yolov8n.pt"

2. You don't need to download official models manually, they will be downloaded automatically.
3. Most likely, you didn't specify LABEL_STUDIO_URL and LABEL_STUDIO_API_KEY in the docker-compose.yml file.
4. Check ML backend logs, the problem should be mentioned there.

Thanks for your reply.

1. The first problem originated from github not label-studio, i pasted the model_path just like above you mentioned, but github added some extra information to it. So its not a problem

2. I also tried not to downloaded official models and still failed though.

3. I have set LABEL_STUDIO_URL as well as LABEL_STUDIO_API_KEY. Because of the reasons we all know of china network, i cant connect to docker-hub freely. So i started ml-backend manually with docker command instead of docker-compose below. And also set LABEL_STUDIO_API_KEY inside the running docker container.

ml-backend starting docker command(followed by docker-compose.yml):
docker run -d --name yolo -p 9090:9090 -e BASIC_AUTH_USER= -e BASIC_AUTH_PASS= -e LOG_LEVEL=DEBUG -e WORKERS=1 -e THREADS=8 -e MODEL_DIR=/data/models -e PYTHONPATH=/app -e LABEL_STUDIO_URL=http://host.docker.internal:8080 -e LABEL_STUDIO_API_KEY= -e ALLOW_CUSTOM_MODEL_PATH=true -e DEBUG_PLOT=false -e MODEL_SCORE_THRESHOLD=0.5 -e MODEL_ROOT=/app/models -v $(pwd)/data/server:/data -v $(pwd)/models:/app/models -v $(pwd)/cache_dir:/app/cache_dir humansignal/yolo:v0
Also the echo result of LABEL_STUDIO_API_KEY inside ml-backend docker container:

4. Can you give me more details on how to get logs in the terminal because of the reason in term 3?

[JasonSloan]

Hi,

1. This line looks wrong in your labeling config

 model_path="[yolov8n.pt](http://yolov8n.pt/)"

it should be

model_path="yolov8n.pt"

2. You don't need to download official models manually, they will be downloaded automatically.
3. Most likely, you didn't specify LABEL_STUDIO_URL and LABEL_STUDIO_API_KEY in the docker-compose.yml file.
4. Check ML backend logs, the problem should be mentioned there.

Thanks i have found the reason why, i should have set LABEL_STUDIO_URL=http://192.168.103.130:8080/ rather than LABEL_STUDIO_URL=host.docker.internal:8080. Now it works!
Thanks a lot for your help!

[JasonSloan]

Sorry, another problem occured. I have to reopen this issue.
Now yolov8 worked well as i could see the boxes predicted by it while yolov11 failed.
I checked the logs and it told me that "Can't get attribute 'C3k2' on <module 'ultralytics.nn.modules.block' from '/opt/conda/lib/python3.10/site-packages/ultralytics/nn/modules/block.py'>".
I tried to import C3k2 in ML-backend env and it seemed that the C3k2 module can be found.
So can you please help me with this?