Hello,
After realizing that my compiled models were not running because I had an SDK mismatch, I am trying to compile artifacts with the right version (SDK 08.04.00.06).
However, I'm facing a problem while compiling the object detection models:
root@1ddb1f2528e5:/home/root/examples/osrt_python/tfl# python3 tflrt_delegate.py -c Running 1 Models - ['tf2_ssd_mobilenet_v2_coco17_ptq'] Running_Model : tf2_ssd_mobilenet_v2_coco17_ptq TIDL Meta PipeLine (Proto) File : Number of OD backbone nodes = 0 Size of odBackboneNodeIds = 0 Warning : concat requires 4D input tensors - only 3 dims present.. Ignore if object detection network Number of subgraphs:3 , 99 nodes delegated out of 102 nodes Warning : concat requires 4D input tensors - only 3 dims present.. Ignore if object detection network ************** Frame index 1 : Running float import ************* **************************************************** ** All the Tensor Dimensions has to be greater then Zero ** DIM Error - For Tensor 95, Dim 3 is -1587263968 **************************************************** TIDL ALLOWLISTING LAYER CHECK: [TIDL_ConcatLayer] Concatenate in batch axis is not supported. WARNING: [TIDL_E_DATAFLOW_INFO_NULL] ti_cnnperfsim.out fails to allocate memory in MSMC. Please look into perfsim log. This model can only be used on PC emulation, it will get fault on target. **************************************************** ** 1 WARNINGS 1 ERRORS ** **************************************************** The soft limit is 2048 The hard limit is 2048 MEM: Init ... !!! MEM: Init ... Done !!! 0.0s: VX_ZONE_INIT:Enabled 0.65s: VX_ZONE_ERROR:Enabled 0.72s: VX_ZONE_WARNING:Enabled 0.26870s: VX_ZONE_INIT:[tivxInit:178] Initialization Done !!! INFORMATION : Standard number of input dimensions is 4, given 3 [1, 1917, 4]-- Will be converted to [1, 1, 1917, 4] (NCHW format) internally INFORMATION : Standard number of input dimensions is 4, given 3 [1, 1917, 91]-- Will be converted to [1, 1, 1917, 91] (NCHW format) internally Warning : Requested Output Data Convert Layer is not Added to the network, It is currently not Optimal Warning : Requested Output Data Convert Layer is not Added to the network, It is currently not Optimal Warning : Requested Output Data Convert Layer is not Added to the network, It is currently not Optimal Warning : Requested Output Data Convert Layer is not Added to the network, It is currently not Optimal ************** Frame index 1 : Running float import ************* WARNING: [TIDL_E_DATAFLOW_INFO_NULL] ti_cnnperfsim.out fails to allocate memory in MSMC. Please look into perfsim log. This model can only be used on PC emulation, it will get fault on target. **************************************************** ** 1 WARNINGS 0 ERRORS ** **************************************************** INFO: Created TensorFlow Lite XNNPACK delegate for CPU. 3.948170s: VX_ZONE_ERROR:[ownTensorCheckSizes:189] Invalid view parameter(s) in dimension: 1
root@1ddb1f2528e5:/home/root/examples/osrt_python/tfl# python3 tflrt_delegate.py -c Running 1 Models - ['ssd_mobilenet_v2_coco_quant_postprocess'] Running_Model : ssd_mobilenet_v2_coco_quant_postprocess TIDL Meta PipeLine (Proto) File : Number of OD backbone nodes = 0 Size of odBackboneNodeIds = 0 Warning : concat requires 4D input tensors - only 3 dims present.. Ignore if object detection network Number of subgraphs:13 , 99 nodes delegated out of 111 nodes ************** Frame index 1 : Running float import ************* **************************************************** ** All the Tensor Dimensions has to be greater then Zero ** DIM Error - For Tensor 96, Dim 2 is 0 **************************************************** WARNING: [TIDL_E_DATAFLOW_INFO_NULL] ti_cnnperfsim.out fails to allocate memory in MSMC. Please look into perfsim log. This model can only be used on PC emulation, it will get fault on target. **************************************************** ** 1 WARNINGS 0 ERRORS ** **************************************************** The soft limit is 2048 The hard limit is 2048 MEM: Init ... !!! MEM: Init ... Done !!! 0.0s: VX_ZONE_INIT:Enabled 0.64s: VX_ZONE_ERROR:Enabled 0.70s: VX_ZONE_WARNING:Enabled 0.22075s: VX_ZONE_INIT:[tivxInit:178] Initialization Done !!! MEM: ERROR: Alloc failed. Invalid size: 0 bytes 0.84757s: VX_ZONE_ERROR:[tivxMemBufferAlloc:113] Memory was not allocated 0.84764s: VX_ZONE_ERROR:[ownAllocTensorBuffer:85] Could not allocate tensor memory 0.84768s: VX_ZONE_ERROR:[ownGraphAllocateDataObject:923] Memory allocation for data reference failed 0.84773s: VX_ZONE_ERROR:[vxVerifyGraph:2000] Memory alloc for data objects failed 0.84776s: VX_ZONE_ERROR:[vxVerifyGraph:2109] Graph verify failed TIDL_RT_OVX: ERROR: Verifying TIDL graph ... Failed !!! TIDL_RT_OVX: ERROR: Verify OpenVX graph failed INFORMATION : Standard number of input dimensions is 4, given 3 [1, 1083, 91]-- Will be converted to [1, 1, 1083, 91] (NCHW format) internally INFORMATION : Standard number of input dimensions is 4, given 3 [1, 600, 91]-- Will be converted to [1, 1, 600, 91] (NCHW format) internally INFORMATION : Standard number of input dimensions is 4, given 3 [1, 150, 91]-- Will be converted to [1, 1, 150, 91] (NCHW format) internally INFORMATION : Standard number of input dimensions is 4, given 3 [1, 54, 91]-- Will be converted to [1, 1, 54, 91] (NCHW format) internally INFORMATION : Standard number of input dimensions is 4, given 3 [1, 24, 91]-- Will be converted to [1, 1, 24, 91] (NCHW format) internally INFORMATION : Standard number of input dimensions is 4, given 3 [1, 6, 91]-- Will be converted to [1, 1, 6, 91] (NCHW format) internally Warning : concat requires 4D input tensors - only 3 dims present.. Ignore if object detection network ************** Frame index 1 : Running float import ************* TIDL ALLOWLISTING LAYER CHECK: [TIDL_ConcatLayer] Concatenate in batch axis is not supported. WARNING: [TIDL_E_DATAFLOW_INFO_NULL] ti_cnnperfsim.out fails to allocate memory in MSMC. Please look into perfsim log. This model can only be used on PC emulation, it will get fault on target. **************************************************** ** 1 WARNINGS 1 ERRORS ** **************************************************** INFORMATION : Standard number of input dimensions is 4, given 3 [1, 1917, 4]-- Will be converted to [1, 1, 1917, 4] (NCHW format) internally INFORMATION : Standard number of input dimensions is 4, given 3 [1, 1917, 91]-- Will be converted to [1, 1, 1917, 91] (NCHW format) internally Warning : Requested Output Data Convert Layer is not Added to the network, It is currently not Optimal Warning : Requested Output Data Convert Layer is not Added to the network, It is currently not Optimal Warning : Requested Output Data Convert Layer is not Added to the network, It is currently not Optimal Warning : Requested Output Data Convert Layer is not Added to the network, It is currently not Optimal ************** Frame index 1 : Running float import ************* WARNING: [TIDL_E_DATAFLOW_INFO_NULL] ti_cnnperfsim.out fails to allocate memory in MSMC. Please look into perfsim log. This model can only be used on PC emulation, it will get fault on target. **************************************************** ** 1 WARNINGS 0 ERRORS ** **************************************************** INFO: Created TensorFlow Lite XNNPACK delegate for CPU. 3.96625s: VX_ZONE_ERROR:[ownTensorCheckSizes:189] Invalid view parameter(s) in dimension: 1 3.96685s: VX_ZONE_ERROR:[tivxUnmapTensorPatch:766] map address is null 3.96692s: VX_ZONE_ERROR:[tivxUnmapTensorPatch:770] map size is equal to 0 3.96971s: VX_ZONE_ERROR:[ownTensorCheckSizes:189] Invalid view parameter(s) in dimension: 1 3.97027s: VX_ZONE_ERROR:[tivxUnmapTensorPatch:766] map address is null 3.97041s: VX_ZONE_ERROR:[tivxUnmapTensorPatch:770] map size is equal to 0
It seems to me that there is an unaccepted operation, but it seems weird because I was able to compile them for SDK 8.6, so I don't understand.
Did you someone encounter this problem?
The other way to go that I have is to flash the newer SDK on the board, but I would like to stay on this one for compatibility with an older project.
