TI Autonomous Driving Algorithms (TIADALG) Library User Guide
tiadalg_image_recursive_nms.h File Reference
#include <stdint.h>

Go to the source code of this file.

Functions

int32_t tiadalg_image_recursive_nms_cn (int8_t *score_buf, int32_t img_width, int32_t img_pitch, int32_t img_height, int32_t cur_score_th, int32_t max_frame_feat, int32_t guard_pixels, int8_t *scratch, int32_t data_type, int32_t *out_buf)
 It does scores buffer thresholding, and then NMS is performed. This is helpful in the scenario where score are sturated in clusters. More...
 
int32_t tiadalg_image_recursive_nms_c66 (int8_t *score_buf, int32_t img_width, int32_t img_pitch, int32_t img_height, int32_t cur_score_th, int32_t max_frame_feat, int32_t guard_pixels, int8_t *scratch, int32_t data_type, int32_t *out_buf)
 It does scores buffer thresholding, and then NMS is performed. This is helpful in the scenario where score are sturated in clusters. More...
 

Function Documentation

◆ tiadalg_image_recursive_nms_cn()

int32_t tiadalg_image_recursive_nms_cn ( int8_t *  score_buf,
int32_t  img_width,
int32_t  img_pitch,
int32_t  img_height,
int32_t  cur_score_th,
int32_t  max_frame_feat,
int32_t  guard_pixels,
int8_t *  scratch,
int32_t  data_type,
int32_t *  out_buf 
)

It does scores buffer thresholding, and then NMS is performed. This is helpful in the scenario where score are sturated in clusters.

tiadalg_image_recursive_nms_cn

Parameters
score_buf[in] : Input 8b score buffer
img_width[in] : Input image width corresponding to score_buf
img_pitch[in] : Input image width pitch corresponding to score_buf
img_height[in] : Input image height corresponding to score_buf
cur_score_th[in] : score threshold value
max_frame_feat[in] : maximum feature points
guard_pixels[in] : guard pixels to be avoided in each direction of x and y.
scratch[in] : scratch buffer
data_type[in] : score_buf data type
out_buf[out] : interleaved x and y of each feature points
Note
  • Step 1 if score_buf[i] < cur_score_th, then score_buf[i] = 0
  • Step 2 Then 3x3 NMS is performed on filtered out scores
  • Size of the buffer scratch is img_width * 2
  • out_buf will have feature points x and y co-ordinated interleaved. x0,y0,x1,y1,x2,y2,x3,y3,... so on
  • return value is total number of generated feature points

◆ tiadalg_image_recursive_nms_c66()

int32_t tiadalg_image_recursive_nms_c66 ( int8_t *  score_buf,
int32_t  img_width,
int32_t  img_pitch,
int32_t  img_height,
int32_t  cur_score_th,
int32_t  max_frame_feat,
int32_t  guard_pixels,
int8_t *  scratch,
int32_t  data_type,
int32_t *  out_buf 
)

It does scores buffer thresholding, and then NMS is performed. This is helpful in the scenario where score are sturated in clusters.

tiadalg_image_recursive_nms_c66

Parameters
score_buf[in] : Input 8b score buffer
img_width[in] : Input image width corresponding to score_buf
img_pitch[in] : Input image width pitch corresponding to score_buf
img_height[in] : Input image height corresponding to score_buf
cur_score_th[in] : score threshold value
max_frame_feat[in] : maximum feature points
guard_pixels[in] : guard pixels to be avoided in each direction of x and y.
scratch[in] : scratch buffer
data_type[in] : score_buf data type
out_buf[out] : interleaved x and y of each feature points
Note
  • Step 1 if score_buf[i] < cur_score_th, then score_buf[i] = 0
  • Step 2 Then 3x3 NMS is performed on filtered out scores
  • Size of the buffer scratch is img_width * 2
  • out_buf will have feature points x and y co-ordinated interleaved. x0,y0,x1,y1,x2,y2,x3,y3,... so on
  • return value is total number of generated feature points

© Copyright 2018 Texas Instruments Incorporated. All rights reserved.
Document generated by doxygen 1.8.6