A Unity ID allows you to buy and/or subscribe to Unity products and services, shop in the Asset Store and participate
in the Unity community.
Hey @WaxyMcRivers I think the allocation you are referring too is a red hearing.
Internally when executing a layer we need to allocate new memory...
So how we optimize for memory allocation is that we first do a pass to figure out what is the largest amount of intermediate layers needed to be...
@cquinteros could you share your onnx model?
We'll happily look into it!
We have a number of yolo architectures that work
Most yolo models there import and work....
@MrOCW I took a look at your model. The problem resides in our importer with batch sizes > 1
Are you able to replace these layers to a...
@laurentlavigne you can check out these projects from Keijiro Takahashi
Hand Pose, Blaze Palm, Blaze Face, Iris tracking
I think there is a confusion with
Hey! No worries at all.
Your code is correct actually.
You are missing the flowing
float outputBuffer = Output.ToReadOnlyArray();
// or slimply...
Ok thanks. What Barracuda version are you on?
Could you clarify a bit what you mean by `does not clean the memory correctly`?
Did you notice memory leaks or high memory usage?
awesome, glad to hear it :)
@michalekmarcin could you provide your model?
My guess would be that it has something to do with the image from the webcam having different...
Looks like tf2onnx does now support exporting TF-2.x to onnx.
I know at one point they were having...
In more details (looking at the end of your network):
AvgPool2D is not a global pool operator. It has a fixed kernel size (in this case...
Yeah testing on 1.3.1
On iPhone X, 719x1280 we got:
Execution of the command buffer was aborted due to an error during execution. Caused GPU Hang...
Second update after checking the issue on iOS. We were able to reproduce it :)
We have some guess to what might be the issue.
So I've tested running the model with different input sizes on a OnePlusA6010 with a GPU worker.
It's not that recent so maybe there is that....
ok thanks a lot. We're looking into it.
@Aurimasp will build a test app for it and we'll investigate.
Initial assumptions looks like a out of...
Could you try using version 1.3.0?
Could you also provide the model you are trying to run? I can test it out on my end to see if I can repro it?