Hi
I have got an onnx model where the input to my network is of size 1x192x256x3. This model is converted to dlc format using the snpe-onnx-to-dlc tool (v 1.51.0.2663). The conversion completes successfully, but when I look at the resulting dlc model the first layers are as follows:
|--------------------------------------------------------------------------------------------------|| Id | Name | Type | Inputs | Outputs | Out Dims | Parameters ||--------------------------------------------------------------------------------------------------|| 0 | input:0 | data | input:0 | input:0 | 1x256x3x192 | input_preprocessing: passthrough || | | | | | | input_type: default || 1 | layer-1 | permute | input:0 | layer-1 | 1x192x256x3 | permute_order: [0, 3, 1, 2] || | | | | | | MACs per inference: 589k (0.139%) || 2 | layer-2 | permute | layer-1 | layer2 | 1x3x192x256 | permute_order: [0, 3, 1, 2] || | | | | | | MACs per inference: 589k (0.139%) || 3 | layer-3 | permute | layer-2 | layer-3 | 1x192x256x3 | permute_order: [0, 2, 3, 1]
So the input dimension is now 1x256x3x192. Then there are three permute layers that to me seem superfluous. The output of the 3rd permute layer is 1x192x256x3, which is exactly the input dimension of my onnx model. How can I get my dlc file to have input dims 1x192x256x3 and get rid of these permute layers?