How many models can I run concurrently? I have 2 models, one is for path segmentation and other is object detection, both in 6 shaves. I have an oak ffc4p.

  • Hi @SOnusingh ,
    this depends on the whole pipeline you want to run. On RVC2 you have 16 shaves in total and these get used by different DepthAI components. For example 3 shaves get used if you use 1080p resolution, 6 if you use 4k resolution, some get used by Stereo node (if you have one), some for ImageManip etc.
    If too many shaves are used (or not optimal amount of shaves is set on NN nodes) you'll get informative logs when you start up your pipeline suggesting you how to configure NN node better.
    If you are working with DepthAIv3 then you have a .superblob file which includes model at all shaves and you can configure this easily by using the
    .setNNArchive(const NNArchive & nnArchive, int numShaves) on the dai.node.NeuralNetwork or dai.node.DetectionNetwork node.
    If you are working with DepthAIv2 though you'll need to go through the export process again and choose different number of shaves.
    Hope this answers your question,
    Klemen

Hi @SOnusingh ,
this depends on the whole pipeline you want to run. On RVC2 you have 16 shaves in total and these get used by different DepthAI components. For example 3 shaves get used if you use 1080p resolution, 6 if you use 4k resolution, some get used by Stereo node (if you have one), some for ImageManip etc.
If too many shaves are used (or not optimal amount of shaves is set on NN nodes) you'll get informative logs when you start up your pipeline suggesting you how to configure NN node better.
If you are working with DepthAIv3 then you have a .superblob file which includes model at all shaves and you can configure this easily by using the
.setNNArchive(const NNArchive & nnArchive, int numShaves) on the dai.node.NeuralNetwork or dai.node.DetectionNetwork node.
If you are working with DepthAIv2 though you'll need to go through the export process again and choose different number of shaves.
Hope this answers your question,
Klemen