hi_mpi_vpc_filter2d

Applicable Products

Product	Supported
Ascend 950PR/Ascend 950DT	√
Atlas A3 training products/Atlas A3 inference products	√
Atlas A2 training products/Atlas A2 inference products	√
Atlas 200I/500 A2 inference products	√
Atlas inference products	☓
Atlas training products	☓

Function

Performs 2D convolutional filtering on an input image.

Prototype

hi_s32 hi_mpi_vpc_filter2d(hi_vpc_chn chn, const hi_filter_2d_param* filter_2d_param, hi_u32* task_id, hi_s32 milli_sec)

Parameters

Parameter	Input/Output	Description
chn	Input	Channel ID for image processing. Ascend 950PR/Ascend 950DT: The value range of this parameter is [0, 256). The maximum number of channels is 256. Atlas A3 training products/Atlas A3 inference products: The value range of this parameter is [0, 256). The maximum number of channels is 256. Atlas A2 training products/Atlas A2 inference products: The value range of this parameter is [0, 256). The maximum number of channels is 256. Atlas 200I/500 A2 inference products: The value range of this parameter is [0, 128). The maximum number of channels is 128.
filter_2d_param	Input	2D convolutional filtering information. The input and output image information and convolutional filtering parameters are configured in the hi_filter_2d_param structure.
milli_sec	Input	Timeout interval, in ms. –1: blocking mode 0: non-blocking mode > 0: timeout interval (timeout mode). The timeout interval varies with the operating system. The deviation is generally within a time slice of an operating system. For example, if the time slice of an operating system is 4 ms and the value of milli_sec is set to 1, the actual timeout interval ranges from 1 ms to 5 ms. When the CPU load is high, the timeout interval may fluctuate.
task_id	Output	Pointer to the task ID, used to identify a task.

Returns

0: success
Other values: failure. For details, see VPC Return Codes.

Restrictions

Model	Restriction
Atlas 200I/500 A2 inference products	The input and output image resolutions must be in the range of [10 × 6, 4096 × 8192]. The input and output images must have the same resolution. The following input image formats are supported (the output image format is the same as the input image format): HI_PIXEL_FORMAT_YUV_400 = 0, // YUV400 8bit HI_PIXEL_FORMAT_YUV_SEMIPLANAR_420 = 1, // YUV420SP NV12 8bit HI_PIXEL_FORMAT_YVU_SEMIPLANAR_420 = 2, // YUV420SP NV21 8bit HI_PIXEL_FORMAT_YUV_SEMIPLANAR_422 = 3, // YUV422SP 8bit HI_PIXEL_FORMAT_YVU_SEMIPLANAR_422 = 4, // YVU422SP 8bit HI_PIXEL_FORMAT_YUYV_PACKED_422 = 7, // YUV422Packed YUYV 8bit HI_PIXEL_FORMAT_YVYU_PACKED_422 = 9, // YUV422Packed YVYU 8bit HI_PIXEL_FORMAT_YUV_PACKED_444 = 11, // YUV444Packed 8bit HI_PIXEL_FORMAT_RGB_888 = 12, // RGB888 HI_PIXEL_FORMAT_BGR_888 = 13, // BGR888 HI_PIXEL_FORMAT_YVU_PACKED_444 = 58, // YVU444Packed 8bit HI_PIXEL_FORMAT_RGB_888_PLANAR = 69, // RGB888 Planar HI_PIXEL_FORMAT_BGR_888_PLANAR = 70, // BGR888 Planar The implementation principle is the same as that of OpenCV. The floating-to-fixed number conversion is different, and the accuracy cannot be exactly same. Due to restrictions on downsampling of YUV images, when the output image format is YUV420SP or YUV422SP, you are advised to set the border making type to border replication mode. This avoids abnormal data on the output image edge.
Ascend 950PR/Ascend 950DT Atlas A3 training products/Atlas A3 inference products Atlas A2 training products/Atlas A2 inference products	The input and output image resolutions must be in the range of [10 × 6, 4096 × 8192]. The input and output images must have the same resolution. The following input image formats are supported (the output image format is the same as the input image format): HI_PIXEL_FORMAT_YUV_400 = 0, // YUV400 8bit HI_PIXEL_FORMAT_YUV_SEMIPLANAR_420 = 1, // YUV420SP NV12 8bit HI_PIXEL_FORMAT_YVU_SEMIPLANAR_420 = 2, // YUV420SP NV21 8bit HI_PIXEL_FORMAT_YUV_SEMIPLANAR_422 = 3, // YUV422SP 8bit HI_PIXEL_FORMAT_YVU_SEMIPLANAR_422 = 4, // YVU422SP 8bit HI_PIXEL_FORMAT_YUYV_PACKED_422 = 7, // YUV422Packed YUYV 8bit HI_PIXEL_FORMAT_YVYU_PACKED_422 = 9, // YUV422Packed YVYU 8bit HI_PIXEL_FORMAT_YUV_PACKED_444 = 11, // YUV444Packed 8bit HI_PIXEL_FORMAT_RGB_888 = 12, // RGB888 HI_PIXEL_FORMAT_BGR_888 = 13, // BGR888 HI_PIXEL_FORMAT_YVU_PACKED_444 = 58, // YVU444Packed 8bit HI_PIXEL_FORMAT_RGB_888_PLANAR = 69, // RGB888 Planar HI_PIXEL_FORMAT_BGR_888_PLANAR = 70, // BGR888 Planar HI_PIXEL_FORMAT_YUV400_FP32 = 138, // YUV400 Packed. Each pixel is represented by a float32 number. HI_PIXEL_FORMAT_RGB_888_PLANAR_FP32 = 1006,// RGB888 Planar. Each pixel is represented by a float32 number. HI_PIXEL_FORMAT_BGR_888_PLANAR_FP32 = 1007,// BGR888 Planar. Each pixel is represented by a float32 number. HI_PIXEL_FORMAT_RGB_888_FP32 = 1012, // RGB888 Packed. Each pixel is represented by a float32 number. HI_PIXEL_FORMAT_BGR_888_FP32 = 1013, // BGR888 Packed. Each pixel is represented by a float32 number. Due to restrictions on downsampling of YUV images, when the output image format is YUV420SP or YUV422SP, you are advised to set the border making type to border replication mode. This avoids abnormal data on the output image edge.

Model

Restriction

Atlas 200I/500 A2 inference products

The input and output image resolutions must be in the range of [10 × 6, 4096 × 8192]. The input and output images must have the same resolution.

The following input image formats are supported (the output image format is the same as the input image format):

HI_PIXEL_FORMAT_YUV_400 = 0,               // YUV400 8bit
HI_PIXEL_FORMAT_YUV_SEMIPLANAR_420 = 1,    // YUV420SP NV12 8bit   
HI_PIXEL_FORMAT_YVU_SEMIPLANAR_420 = 2,    // YUV420SP NV21 8bit 
HI_PIXEL_FORMAT_YUV_SEMIPLANAR_422 = 3,    // YUV422SP 8bit
HI_PIXEL_FORMAT_YVU_SEMIPLANAR_422 = 4,    // YVU422SP 8bit
HI_PIXEL_FORMAT_YUYV_PACKED_422 = 7,       // YUV422Packed YUYV 8bit
HI_PIXEL_FORMAT_YVYU_PACKED_422 = 9,       // YUV422Packed YVYU 8bit
HI_PIXEL_FORMAT_YUV_PACKED_444 = 11,      // YUV444Packed  8bit
HI_PIXEL_FORMAT_RGB_888 = 12,             // RGB888
HI_PIXEL_FORMAT_BGR_888 = 13,             // BGR888
HI_PIXEL_FORMAT_YVU_PACKED_444 = 58,       // YVU444Packed 8bit
HI_PIXEL_FORMAT_RGB_888_PLANAR = 69,       // RGB888 Planar
HI_PIXEL_FORMAT_BGR_888_PLANAR = 70,       // BGR888 Planar

The implementation principle is the same as that of OpenCV. The floating-to-fixed number conversion is different, and the accuracy cannot be exactly same.
Due to restrictions on downsampling of YUV images, when the output image format is YUV420SP or YUV422SP, you are advised to set the border making type to border replication mode. This avoids abnormal data on the output image edge.

Ascend 950PR/Ascend 950DT

Atlas A3 training products/Atlas A3 inference products

Atlas A2 training products/Atlas A2 inference products

The input and output image resolutions must be in the range of [10 × 6, 4096 × 8192]. The input and output images must have the same resolution.

The following input image formats are supported (the output image format is the same as the input image format):

HI_PIXEL_FORMAT_YUV_400 = 0,               // YUV400 8bit
HI_PIXEL_FORMAT_YUV_SEMIPLANAR_420 = 1,    // YUV420SP NV12 8bit   
HI_PIXEL_FORMAT_YVU_SEMIPLANAR_420 = 2,    // YUV420SP NV21 8bit 
HI_PIXEL_FORMAT_YUV_SEMIPLANAR_422 = 3,    // YUV422SP 8bit
HI_PIXEL_FORMAT_YVU_SEMIPLANAR_422 = 4,    // YVU422SP 8bit
HI_PIXEL_FORMAT_YUYV_PACKED_422 = 7,       // YUV422Packed YUYV 8bit
HI_PIXEL_FORMAT_YVYU_PACKED_422 = 9,       // YUV422Packed YVYU 8bit
HI_PIXEL_FORMAT_YUV_PACKED_444 = 11,      // YUV444Packed  8bit
HI_PIXEL_FORMAT_RGB_888 = 12,             // RGB888
HI_PIXEL_FORMAT_BGR_888 = 13,             // BGR888
HI_PIXEL_FORMAT_YVU_PACKED_444 = 58,       // YVU444Packed 8bit
HI_PIXEL_FORMAT_RGB_888_PLANAR = 69,       // RGB888 Planar
HI_PIXEL_FORMAT_BGR_888_PLANAR = 70,       // BGR888 Planar
HI_PIXEL_FORMAT_YUV400_FP32 = 138,         // YUV400 Packed. Each pixel is represented by a float32 number.
HI_PIXEL_FORMAT_RGB_888_PLANAR_FP32 = 1006,// RGB888 Planar. Each pixel is represented by a float32 number.
HI_PIXEL_FORMAT_BGR_888_PLANAR_FP32 = 1007,// BGR888 Planar. Each pixel is represented by a float32 number.
HI_PIXEL_FORMAT_RGB_888_FP32 = 1012,       // RGB888 Packed. Each pixel is represented by a float32 number.
HI_PIXEL_FORMAT_BGR_888_FP32 = 1013,       // BGR888 Packed. Each pixel is represented by a float32 number.

Due to restrictions on downsampling of YUV images, when the output image format is YUV420SP or YUV422SP, you are advised to set the border making type to border replication mode. This avoids abnormal data on the output image edge.

Parent topic: VPC