1) Hog Cells. The format of both output tensors is VX_TYPE_INT16. But there is no q7.8 translation, i.e. multiplication by powf(2, 8). This leads to zero featurestensor in Hog Features.
2) Hog Features. There is no window [processing], i.e. window_width, window_height, window_stride and block_stride are not used for features tensor computation.
@kiritigowda
Output of Hog Cells are tensors of VX_TYPE_INT16 with fixed point 8.
But here is converting from a floating-point value to a fixed-point value without using this fixed point.
Output tensors' values don't meet standard.
1) Hog Cells. The format of both output tensors is
VX_TYPE_INT16
. But there is noq7.8
translation, i.e. multiplication by powf(2, 8). This leads to zerofeatures
tensor in Hog Features.2) Hog Features. There is no window [processing], i.e.
window_width
,window_height
,window_stride
andblock_stride
are not used forfeatures
tensor computation.