search

ceccocats / tkDNN

Deep neural network library and toolkit to do high performace inference on NVIDIA jetson platforms
GNU General Public License v2.0
717 stars 209 forks source link

Yolov4tiny engine build error #118

Closed ahmedius2 closed 3 years ago

ahmedius2 commented 4 years ago

Hi! Firstly, thanks for sharing this nice project with us.

I am trying to run YOLOv4-tiny on NVDLA of Jetson AGX. Since leaky relu and mish activations are not supported on NVDLA, I changed all leaky and mish layers into relu in the darknet .cfg files, like this: activation=leaky --> activation=relu activation=mish --> activation=relu For the YOLOv4, YOLOv3, and YOLOv3-tiny, it builds the CUDA engine and serialize it without any problem and all activation layers become supported on DLA. However, the engine does not build on YOLOv4-tiny, the program aborts by saying could not build the engine. What could be the reason for this?

Best regards,

Ahmet Soyyigit

mive93 commented 4 years ago

Hi @ahmedius2

It's indeed weird, could you please activate the debug (check README ) and copy here the output please?

mive93 commented 3 years ago

Closing for now, feel free to reopen.