I

izou3

xperroni

Got it. I used this tool here to convert it to onnx and passed that through the blob which seemed to work too.

Thanks for your help!

jakaskerl

Yes usually before I use tools.luxonis but it wasn't working yesterday as it seems the server is down. I tried it today and everything works fine. Note that I also was able to solve this by using luxonis tools to convert it to onnx and passing that onnx version to the blob converter.

Thanks for your help!

Hello,

When I run my fine tuned yolov8 detection model on an Oakd SR, it outputs an error

Mask is not defined for output layer with width '3549'. Define at pipeline build time using 'setAnchorMasks' for 'side3549'

My understanding is that yolov8 is anchorless so I am not sure why it is outputting this error. When I try using the blob model provided by Luxonis (yolov8n_coco_640x352.blob), there isn't any issues.

I tried setting anchors for the DetectionNetwork but it is still an issue. Note that I am using BlobConverter to convert my fine tuned yolov8n model. I tried using tools.luxonis but it keeps outputing an error 500.

Can anyone provide an insight into my it is outputting that error even though you are not suppose to specify anchors for yolov8?

jakaskerl Got it. I will give both options a try

I have an OakdSR pointing down at a conveyor belt at a distance of 16 inches. It is generating the depth maps at a frame rate of 5 FPS with the following configuration settings attached below.

However the depth map and the resulting point cloud are very inaccurate. I have posted some sample pictures below. It is shaped in a pyramid-like structure with different layers for different height even though the entire scene should be at one height as the camera is pointing down at a flat surface. I have read the intrinsic matrix from the device systems based on the image resolution to generate the point cloud. However I do not believe that this is an issue with the point cloud generation as the depth map has different layers of depth distances. I would appreciate any advice or help on what the issue is and how to make the depth map more accurate. Thanks!

This is a link to the original RGB image, depth map, and Point Cloud visualization in a drive

Drive Link

Stereo Configuration

stereo->setDefaultProfilePreset(dai::node::StereoDepth::PresetMode::HIGH_ACCURACY);
    stereo->setLeftRightCheck(true); // LR-check is required for depth alignment
    stereo->setExtendedDisparity(true);
    stereo->setSubpixel(true);
    stereo->setDepthAlign(dai::CameraBoardSocket::CAM_C);
   
    auto config = stereo->initialConfig.get();
    config.postProcessing.speckleFilter.enable = false;
    config.postProcessing.speckleFilter.speckleRange = 50;
    config.postProcessing.temporalFilter.enable = false;
    config.postProcessing.spatialFilter.enable = true;
    config.postProcessing.spatialFilter.holeFillingRadius = 2;
    config.postProcessing.spatialFilter.numIterations = 1;
    config.postProcessing.thresholdFilter.minRange = 300;
    config.postProcessing.thresholdFilter.maxRange = 600;
    config.postProcessing.decimationFilter.decimationFactor = 1;
    stereo->initialConfig.set(config);