In Image_segmentation task, the generated .ptl file by _save_for_lite_interpreter produces a model that outputs a [1,1000] tensor instead of a DICT_STRING_KEY. In the android source, if we replace the generated file with the file before optimization, the application works correctly.
In Image_segmentation task, the generated
.ptl
file by_save_for_lite_interpreter
produces a model that outputs a [1,1000] tensor instead of aDICT_STRING_KEY
. In the android source, if we replace the generated file with the file before optimization, the application works correctly.