Open rahulsharma11 opened 5 years ago
Hi, I know the team might be having very busy schedule to rectify the issues here. It will be greate if you can provide some solution/reason regarding my above querry. The models are from Baidu website snf test code is default in Tengine. Then what can be the issue? Is it rockpro specific? or something else.
Is there any change in "tf_lite_mssd.cpp" needs to be done? And why res tf_lite examples are not running on my rockpro64? Will appreciate your time.
Thanks.
Hi, I used latest Tengine with ComputeLibrary and enabled TFLITE_SERIALIZER. I downloaded the tflte models from baidu website and placed in the respective "models" folder. I got following resutls with different test code- 1) With ./build/test/bin/tf_lite_mssd- It took around 13 seconds to forward pass.
2) With ./build/tests/bin/tf_lite_mssd_quant- ACL Graph Initialized Driver: ACLGraph probed 1 devices create graph done! cpu_runner.cpp:759 graph: models/detect.tflite active num is: 4 output nodes: 1 graph_task.cpp:234 failed to Prerun task on dev executor: arm.probed prerun _graph failed
3) With ./build/tests/bin/tf_lite_mobilenet_quant ACL Graph Initialized Driver: ACLGraph probed 1 devices create graph done! graph_task.cpp:234 failed to Prerun task on dev executor: arm.probed prerun _graph failed
Any suggestions? why above test codes are not working. Anything else need to be done? Thanks.