doannguyentrong/object-detection-tflite-cpp

#object detection #rb5 #qualcomm #hexagon #gpu #inference

object-detection-tflite-cpp's Issues

RB5 Hexagon DSP

I am working on rb5, detection model for integrate to Qt Creator (c++). I just want to increase FPS of video that detected. Also I am using aarch64 not android. Which shared library should I add? Is it necessary libhexagon_nn_skel library?

Error when compile repo object-detection-tflite-cpp

Hi Doan,
I cloned repo object-detection-tflite-cpp to my host machine (Laptop ubuntu 18). When I run ./compile.sh command, there is an error:

If I clone and run on RB5 device, there is also an error:

Could help me how to fix those errors? Thank you very much!

OP is supported, but tensor type isn't matched!

Namaste,

I am not able to run the detection on GPU of RB5 processor. When I am running the detector (I am using a video file as input), I am getting following message:

sh-5.0# ./object-detector 0 ../test_for_rb5.mp4
GPU Delegate!
Video source: ../test_for_rb5.mp4
EXISTS: models/default/detect.tflite
EXISTS: models/default/labelmap.txt
Loading: models/default/labelmap.txt
Done!!
Loading: models/default/detect.tflite
Done!!
WARNING: Created TensorFlow Lite XNNPACK delegate for CPU.
Done!!
Quantized model! Can use Hexagon Delegate!
Enabling GPU Delegate!
INFO: Created TensorFlow Lite delegate for GPU.
: Following operations are not supported by GPU delegate:
CONCATENATION: OP is supported, but tensor type isn't matched!
CONV_2D: OP is supported, but tensor type isn't matched!
CUSTOM TFLite_Detection_PostProcess: TFLite_Detection_PostProcess
DEPTHWISE_CONV_2D: OP is supported, but tensor type isn't matched!
LOGISTIC: OP is supported, but tensor type isn't matched!
RESHAPE: OP is supported, but tensor type isn't matched!
No operations will run on the GPU, and all 64 operations will run on the CPU.
INFO: Created 0 GPU delegate kernels.

Thanks in advance for any pointers.