I'm trying to use vision_apps to demo the deep learning model with custom layer on PC emulation.
I've modified the sample which is named app_tidl. (remove the diplayOutput function)
I've already done some things as below:
- The model with custom layer can import and inference normally.
- modified related .mak files :
-
- psdk_rtos_auto_j7_06_02_00_21\tiovx\psdkra_tools_path.mak
+ TIDL_CUSTOM_PATH ?= $(PSDK_PATH)/tidl_j7_01_00_00_10/ti_dl/custom - psdk_rtos_auto_j7_06_02_00_21\tiovx\kernels_j7\concerto_inc.mak
+ STATIC_LIBS += tidl_custom
+ LDIRS+= $(PSDK_PATH)/tidl_j7_01_01_00_10/ti_dl/lib/PC/dsp/algo/$(TARGET_BUILD) - psdk_rtos_auto_j7_06_02_00_21\vision_apps\apps\concerto_x86_64_inc.mak
+ STATIC_LIBS += tidl_custom
+ LDIRS+= $(PSDK_PATH)/tidl_j7_01_01_00_10/ti_dl/lib/PC/dsp/algo/$(TARGET_BUILD) - "Required system library libtidl_custom.a" shows in command window.
- psdk_rtos_auto_j7_06_02_00_21\tiovx\psdkra_tools_path.mak
The program is crash in vxProcessGraph function in main.c.
But it can work normally when model remove custom layer.
So my questions are :
- Do "Required system library libtidl_custom.a" means the link is succeeded? If not, how to check the result?
- How to demo the model with custom layer at vision_apps/apps/dl_demo/app_tidl or demo in other ways.
There are some relative files as below (custom folder in tidl_j7_01_01_00_10/ and model):