Bug in VLIB_xyGradients_Magnitude_Orientations

Li Zhang2

Hi,

I'm using vlib_c66x_3_1_0_9.

I think there is a bug in VLIB_xyGradients_Magnitude_Orientations like the previously fixed VLIB_integralImage8, the function can't handle image with more than 2^16 pixels. The bug only resides in the optimized version, *_cn version is clean.

Here are some test images,

1) magnitude output from optimized version

2) magnitude output from natural C version

The first output gets cut at some row. The image size is 384x288, from the row that the problem start to rise, it's about 2^16 pixels. The same problem exists in the output of orientation.

And same problem spotted in older version like vlib 3.0.0.8.

Can anyone confirm this problem? Thanks.

Regards,

over 9 years ago

0 Raja over 9 years ago

TI__Guru* 81335 points

Hi Li,

We will get back to you shortly. Thank you for the post.

0 Sudhakar Ayyasamy over 9 years ago in reply to Raja

TI__Intellectual 1010 points

Li,

Thank you for your patience. We confirm that it is a bug, and it will be fixed in the next VLIB release. Thank you for pointing it out.

Sincerely,

Sudhakar Ayyasamy

0 zhongqing ouyang over 8 years ago in reply to Sudhakar Ayyasamy

Prodigy 100 points

When I use the function:

void VLIB_goodFeaturestoTrack	(	const uint16_t	im[restrict],
		uint8_t	out[restrict],
		int32_t	width,
		int32_t	height,
		int16_t	thresh,
		uint8_t	patch,
		uint8_t	qualitypar,
		int16_t	topNfeatures,
		uint8_t	mindist,
		uint16_t	outTemp[],
		int32_t *restrict	good_points_number,
		int16_t	pixIndex[restrict],
		uint16_t	internalBuf[restrict],
		int32_t	ind[restrict]
	)

when I input image's width is 640, and height is 480, the output out is mistake. only width*height is 160*120, the result out is correct. can I get the Source code?

0 zhongqing ouyang over 8 years ago in reply to zhongqing ouyang

Prodigy 100 points

And i'm use vlib_c674x_3_2_0_2 and vlib_c674x_3_1_0_9.

Processors

Processors forum

Bug in VLIB_xyGradients_Magnitude_Orientations