对Yolov8n.pt转wts后，再序列化构建时，addResize()函数返回值为空，应如何解决？

[retioa11]

Tasks

Beta Give feedback

No tasks being tracked yet.

[retioa11]

想把这个仓库的代码，[https://github.com/wang-xinyu/tensorrtx/tree/master/yolov8，修改为支持推理时动态形状，如下图的红框所示：

修改了红框里的三四行代码，第100行就返回为空了，该如何解决呢？

[lix19937]

You can upload the code for the two versions before and after here. @retioa11

[retioa11]

You can upload the code for the two versions before and after here. @retioa11
different：

code for the two versions before and after：
[
yolov8_before.zip
](url)
yolov8_after.zip
before：

after：

[lix19937]

Are your H,W are dynamic shape ?

If H W are dynamic shape, you can ref https://github.com/lix19937/trt-samples-for-hackathon-cn/blob/master/cookbook/02-API/OptimizationProfile/main-TensorInput.py#L28

[retioa11]

Oh, thank you. But wondering why my modification is returned empty and how should be modified to serialize Yolov8. Does the code in the link help? Not very understand.

[lix19937]

If you use dynamic shape, you should set dynamic binding's shape at each infer.
Like https://github.com/lix19937/trt-samples-for-hackathon-cn/blob/master/cookbook/02-API/OptimizationProfile/main-TensorInput.py#L50
and
https://github.com/NVIDIA/TensorRT/blob/release/10.1/samples/sampleDynamicReshape/sampleDynamicReshape.cpp#L428-L432

[lix19937]

As usual, it's caused by not setting the right dynamic shape profile or rt-input shapes.

[retioa11]

Thanks, I'll have it a try。