quantize hta model, run below command
snpe-dlc-quantize --input_dlc espcn_250.dlc --input_list ../data/250_image_file_list.txt --output_dlc espcn_250_quantized_hta.dlc --enable_hta
[INFO] InitializeStderr: DebugLog initialized.
[INFO] Writing intermediate model
[WARNING] NetworkTopology::populateNetworkDesc network desc inputs is empty. Does this network have data/input layer?
[AIP_TF8_HTA : 0 1 2 3 4 5 ] --> [DSP_TF8 : 6 ] ::2
ESPCN/DepthToSpace: PixelShuffle not supported
[INFO] Setting activation for layer: input:0 and buffer: input:0
[INFO] bw: 8, min: 0.000000, max: 253.000000, delta: 0.992157, offset: 0.000000
[INFO] Setting activation for layer: ESPCN/Conv/Conv2D and buffer: ESPCN/Conv/BiasAdd:0
[INFO] bw: 8, min: -126.273250, max: 177.497304, delta: 1.191257, offset: -106.000000
[INFO] Setting activation for layer: ESPCN/Conv/Tanh and buffer: ESPCN/Conv/Tanh:0
[INFO] bw: 8, min: -1.003922, max: 0.996078, delta: 0.007843, offset: -128.000000
[INFO] Setting activation for layer: ESPCN/Conv_1/Conv2D and buffer: ESPCN/Conv_1/BiasAdd:0
[INFO] bw: 8, min: -3.806942, max: 4.215952, delta: 0.031462, offset: -121.000000
[INFO] Setting activation for layer: ESPCN/Conv_1/Tanh and buffer: ESPCN/Conv_1/Tanh:0
[INFO] bw: 8, min: -0.995378, max: 1.003216, delta: 0.007838, offset: -127.000000
[INFO] Setting activation for layer: ESPCN/Conv_2/Conv2D and buffer: ESPCN/Conv_2/BiasAdd:0
[INFO] bw: 8, min: -1.239131, max: 3.776398, delta: 0.019669, offset: -63.000000
[INFO] Setting activation for layer: ESPCN/DepthToSpace and buffer: ESPCN/DepthToSpace:0
[INFO] bw: 8, min: -1.239131, max: 3.776398, delta: 0.019669, offset: -63.000000
[INFO] Running Graph Partitioner for SDM865
[INFO] Blob ID:2
[WARNING] Dropping subnet 0: <0, 1> to improve runtime performance.
[WARNING] Dropping subnet 1: <3, 3> to improve runtime performance.
[WARNING] Dropping subnet 2: <5, 5> to improve runtime performance.
[INFO] Writing quantized model to: espcn_250_quantized_hta.dlc
[INFO] Compiling HTA metadata into DLC.
ESPCN/DepthToSpace: PixelShuffle not supported
[INFO] Creating new AIP record aip.metadata0
[INFO] Record Version:: 1.2.0.0
[INFO] Compiler Version:: 1.6.1.4
[INFO] HTA Blob ID:: 1
[WARNING] No manual partitions specified. Resorting to automatic partitioning.
[INFO] Received HTA subnet 0: <0, 1>
[WARNING] Dropping subnet 0: <0, 1> to improve runtime performance.
[INFO] Received HTA subnet 1: <3, 3>
[WARNING] Dropping subnet 1: <3, 3> to improve runtime performance.
[INFO] Received HTA subnet 2: <5, 5>
[WARNING] Dropping subnet 2: <5, 5> to improve runtime performance.
[INFO] Creating new AIP record aip.metadata1
[INFO] Record Version:: 1.2.0.0
[INFO] Compiler Version:: 1.6.1.4
[INFO] HTA Blob ID:: 2
[INFO] Successfully compiled HTA metadata into DLC.
[INFO] DebugLog shutting down.
error:
ESPCN/DepthToSpace: PixelShuffle not supported
but in Supported Network Layers.
DepthToSpace is supported in DSP mode, why is not supported?
in my case, pixelshuffle also not supported in gpu mode. While convertting onnx model to dlc model, snpe will split pixelshuffle into several operations in which some operations not supported in gpu mode, for example reshpe will exceed the limitation. Still looking for solutions.