TDA4VM: numBatches = 6，模型转化不成功

还有一个问题：（模型都有相同的输入：288*512）在模型转化时，设置numBatches = 6，为什么resnet18和resnet34可以正确转化，而resnet50转化不成功呢

您好，您的问题我们需要升级到英文论坛寻求帮助，链接如下，有答复尽快给您：

https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1283143/tda4vm-numbatches-6-model-conversion-failed

您好，

请问您用的SDK版本是多少？从您分享的log来看似乎是8.2。

从9.0 SDK 起，对多核 DSP AM69A/TDA4VH 提供batch processing支持。 

在更换SDK9.0后，出现图中问题，应该是少了步骤，请问有最新的TI Deep Learning Product User Guide吗？

是否支持Ubuntu18.04

您好，

升级SDK到9.0后仍不支持大尺寸多batch（1024*576,batch4输入）模型在板端的推理吗？WARNING: [TIDL_E_DATAFLOW_INFO_NULL] Network compiler returned with error or didn't executed, this model can only be used on PC/Host emulation mode, it is not expected to work on target/EVM.

执行如下操作：

gmm@gmm:~/tda4/ti-processor-sdk-rtos-j721e-evm-09_00_00_02/c7x-mma-tidl/ti_dl/utils/tidlModelImport$ ./out/tidl_model_import.out /home/gmm/tda4/ti-processor-sdk-rtos-j721e-evm-09_00_00_02/c7x-mma-tidl/ti_dl/test/testvecs/config/import/public/onnx/tidl_import_batch4_resnet18_1024_576.txt
ONNX Model (Proto) File : ../../test/testvecs/models/public/onnx/resnet18-1024_576-sim.onnx
TIDL Network File : ../../test/testvecs/config/tidl_models/onnx/tidl_net_batch4_resnet18_1024.bin
TIDL IO Info File : ../../test/testvecs/config/tidl_models/onnx/tidl_io_batch4_resnet18_1024_
Current ONNX OpSet Version : 8
Empty prototxt path, running calibration

~~~~~Running TIDL in PC emulation mode to collect Activations range for each layer~~~~~

Processing config file #0 : /home/gmm/tda4/ti-processor-sdk-rtos-j721e-evm-09_00_00_02/c7x-mma-tidl/ti_dl/test/testvecs/config/tidl_models/onnx/tidl_import_batch4_resnet18_1024_576.txt.qunat_stats_config.txt
Freeing memory for user provided Net
----------------------- TIDL Process with REF_ONLY FLOW ------------------------

# 0 . .. T 264461.15 .... ..... ...
A : 895, 0.0000, 1.0000, 405 .... .....
# 1 . .. T 271820.34 .... ..... ...
A : 557, 0.0000, 1.0000, 733 .... .....
# 2 . .. T 264794.91 .... ..... ...
A : 442, 0.0000, 0.6667, 675 .... .....
# 3 . .. T 263599.49 .... ..... ...
A : 498, 0.0000, 0.5000, 456 .... .....


***************** Calibration iteration number 0 started ************************

~~~~~Running TIDL in PC emulation mode to collect Activations range for each layer~~~~~

Processing config file #0 : /home/gmm/tda4/ti-processor-sdk-rtos-j721e-evm-09_00_00_02/c7x-mma-tidl/ti_dl/test/testvecs/config/tidl_models/onnx/tidl_import_batch4_resnet18_1024_576.txt.qunat_stats_config.txt
Freeing memory for user provided Net
----------------------- TIDL Process with REF_ONLY FLOW ------------------------

# 0 . .. T 223584.37 .... ..... ...
A : 895, 0.0000, 1.0000, 405 .... .....
# 1 . .. T 222254.42 .... ..... ...
A : 557, 0.0000, 0.5000, 733 .... .....
# 2 . .. T 222516.39 .... ..... ...
A : 442, 0.0000, 0.3333, 656 .... .....
# 3 . .. T 221994.04 .... ..... ...
A : 498, 0.0000, 0.2500, 475 .... .....


***************** Calibration iteration number 0 completed ************************

------------------ Network Compiler Traces -----------------------------
successful Memory allocation
Floating point exception (core dumped)
Could not open /home/gmm/tda4/ti-processor-sdk-rtos-j721e-evm-09_00_00_02/c7x-mma-tidl/ti_dl/test/testvecs/config/tidl_models/onnx/batch4_resnet18_1024/perfSimInfo.bin
WARNING: [TIDL_E_DATAFLOW_INFO_NULL] Network compiler returned with error or didn't executed, this model can only be used on PC/Host emulation mode, it is not expected to work on target/EVM.
****************************************************
** 1 WARNINGS 0 ERRORS **
****************************************************

我们转达给工程师看下。

您好，

非常抱歉回复晚了。

您使用的是哪个特定于 SoC 的 SDK？

正如之前提到的，multicore器件(TDA4VH/AM69A)支持批batch processing，请确保使用了同样的。

此外请注意：

Number of batches <= Number of cores

对于 TDA4VH/AM69A，DSP 内核的最大数量为4个，因此支持batch processing大小为4的型号。 通过在4个批次模型上迭代运行推理，可以实现更多批次。