search

ARM-software / armnn

Arm NN ML Software. The code here is a read-only mirror of https://review.mlplatform.org/admin/repos/ml/armnn
https://developer.arm.com/products/processors/machine-learning/arm-nn
MIT License
1.14k stars 307 forks source link

Pooling2dLayer: TensorShape set on OutputSlot[0] does not match the inferred shape #641

Closed xiaotongnii closed 1 year ago

xiaotongnii commented 2 years ago

1.env: armnn21.02 tensorflow:2.3.1 model:mobilenet_v1_1.0_224_frozen.pb model:NHWC, batch,224,224,3

2.When inference tensorflw model on armnn,LayerValidationException Happend

load network pass terminate called after throwing an instance of 'armnn::LayerValidationException' what(): Pooling2dLayer: TensorShape set on OutputSlot[0] does not match the inferred shape. : [4,2147483648,1,1024] != [4,2147483649,1,1024] Aborted (core dumped)

MikeJKelly commented 2 years ago

Hi @Shelton-N

There seems to be something wrong with your model: [4,2147483648,1,1024] != [4,2147483649,1,1024]

That second dimension is far too large for a TF Mobilenet model but the assertion is correct as 2147483648 is not equal to 2147483649.

The output of that Pooling2D would require 8192 GB of memory.

Support for TensorFlow was dropped in 21.02 I'd recommend converting the model to TFLite and using a more up to date version of ArmNN.

Best regards, Mike

xiaotongnii commented 1 year ago

Hi @Shelton-N

There seems to be something wrong with your model: [4,2147483648,1,1024] != [4,2147483649,1,1024]

That second dimension is far too large for a TF Mobilenet model but the assertion is correct as 2147483648 is not equal to 2147483649.

The output of that Pooling2D would require 8192 GB of memory.

Support for TensorFlow was dropped in 21.02 I'd recommend converting the model to TFLite and using a more up to date version of ArmNN.

Best regards, Mike

@MikeJKelly OK，I use armnn22.02 to deploy TFLite model ，everything is right。