Testing Yolov26s on BI6
- Thread starter Tinman
- Start date
Interesting. I've been following the other (YOLO8 vs. YOLO10) thread where another user mentioned this YOLO26 model and gave some hints on downloading direct from Ultralytics and converting the format, but perhaps you could elaborate or provide a means to directly download the model(s) you are using.
I suspect most of us will be quite envious of your detection times, but it's not surprising given the NVIDIA card you are using: frankly I'm surprised you even detected a difference from other models. What did it shave off, maybe 1-2ms?
I suspect most of us will be quite envious of your detection times, but it's not surprising given the NVIDIA card you are using: frankly I'm surprised you even detected a difference from other models. What did it shave off, maybe 1-2ms?
Last edited:
I can give you some more detail regarding downloading the model...
The model can be found at YOLO26s Model by Ultralytics
You'll see tabs right below the charts, one of which is Export. Click on that and you'll be prompted to create a free account. Once you've done that (takes less than a minute) you'll be able to download the onnx version of the file. There's nothing required for the conversion to onnx, just select the export default values and download. Be sure to rename it to yolov26s.onnx and copy it to whatever directory that the AImodelpack.exe extracted to when you install it. Restart Blueiris and you're ready to select it.
The model can be found at YOLO26s Model by Ultralytics
You'll see tabs right below the charts, one of which is Export. Click on that and you'll be prompted to create a free account. Once you've done that (takes less than a minute) you'll be able to download the onnx version of the file. There's nothing required for the conversion to onnx, just select the export default values and download. Be sure to rename it to yolov26s.onnx and copy it to whatever directory that the AImodelpack.exe extracted to when you install it. Restart Blueiris and you're ready to select it.
Yes, maybe 3-5ms, but so many variables in conditions make that hard to say. The accuracy did go from the 80s to 90s% though. That is exactly where I dl the model. I just logged in as my github ID and you then add whatever model to your project and then you can export the model in the onnx format.
Here:
Train and Deploy Ultralytics YOLO Models
End-to-end computer vision platform with 125K+ GitHub stars. Train Ultralytics YOLO models, manage datasets, and deploy with one click.
platform.ultralytics.com
Here are the small and medium models :
Last edited:
Seems ok
I don't have .dat analysis on this cam, but it detected the cat fine and this is a very dark area.
Last edited:
yes, I am currently using yolov26s for my primary model and it seems more accurate and faster than bi-combinedv5. I wanted to test the yolov26m on 2 of my cams running those using the "additional models" but cannot yet. I hope Ken will get that "add new model" button fixed soon.