Revision history [back]

The detectMultiScale function automatically uses a sliding window over the whole image, and inside such windows the HOG descriptor is computed, with the corresponding set cell and block parameters. The win_stride parameter controls the movement of the sliding window. A big stride will cause consecutive positions of the sliding window to be far apart (no overlapping) while small stride will cause to have a big overlapped region between two consecutive positions of the sliding window. Obviously, using smaller stride the function will perform more computations (will check more regions), which will likely lead to more precision while also requiring more computations.

So, the win_stride parameter is not strictly related to HOG itself but with the method to go over the image. The parameter block_stride is however an internal parameter for HOG computation (if you've read that literature, you've seen it is used to move the blocks over which the normalization of cells is done)

The detectMultiScale function automatically uses a sliding window over the whole image, and inside such windows the HOG descriptor is computed, with the corresponding set cell and block parameters. The win_stride parameter controls the movement of the sliding window. A big stride will cause consecutive positions of the sliding window to be far apart (no overlapping) while small stride will cause to have a big overlapped region between two consecutive positions of the sliding window. Obviously, using smaller stride the function will perform more computations (will check more regions), which but it will likely lead to more precision while also requiring more computations.of detections.

So, the win_stride parameter is not strictly related to HOG itself but with the method to go over the image. The parameter block_stride is however an internal parameter for HOG computation (if you've read that literature, you've seen it is used to move the blocks over which the normalization of cells is done)