Gst-nvvideoconvert#

This plugin performs video color format conversion. It accepts NVMM memory as well as RAW (memory allocated using calloc() or malloc() ), and provides NVMM or RAW memory at the output.

Inputs and Outputs#

Inputs
- Gst Buffer batched buffer
- NvDsBatchMeta
- Format: (NVMM/RAW)
- dGPU - NV12, I420, P010_10LE, BGRx, RGBA, GRAY8, RGB, BGR, Y444, Y444_10LE, Y444_12LE, BGR10A2_LE, UYVP, I420_12LE, RGB10A2_LE, GRAY16_LE
- Jetson - NV12, I420, P010_10LE, BGRx, RGBA, GRAY8, RGB, BGR, BGR10A2_LE, UYVP, UYVY, YUY2, YVYU, Y42B, I420_12LE, GRAY16_LE
Control parameters
- gpu-id (dGPU only)
- nvbuf-memory-type
- src-crop
- dst-crop
- interpolation-method
- compute-hw
- flip-method
Output
- Gst Buffer
- NvDsBatchMeta
- Format: (NVMM/RAW)
- dGPU - NV12, I420, P010_10LE, BGRx, RGBA, GRAY8, RGB, BGR, Y444, Y444_10LE, Y444_12LE, BGR10A2_LE, UYVP, I420_12LE, RGB10A2_LE, GRAY16_LE
- Jetson - NV12, I420, P010_10LE, BGRx, RGBA, GRAY8, RGB, BGR, BGR10A2_LE, UYVP, UYVY, YUY2, YVYU, Y42B, I420_12LE, GRAY16_LE

Note

For 12bit NV12 format, the input/output caps shows I420_12LE. But the data is handled as semiplanar 12 bit data. This is a workaround since gstreamer v1.16 does not have the particular caps for 12 bit NV12. Downstream component should process it considering as 12bit NV12 only.
Only NVMM memory support is added for I420_12LE, Y444_10LE and Y444_12LE formats.

Features#

This plugin supports scaling and conversion for NVMM to NVMM, RAW to NVMM, and NVMM to RAW and RAW to RAW buffer type conversion. The plugin supports cropping of the input and output frames.

Gst Properties#

The following table describes the Gst properties of the Gst-nvvideoconvert plugin.

Gst-nvvideoconvert gst properties#

Property

Meaning

Type and Range

Example Notes

nvbuf-memory-type

Type of memory to be allocated. For dGPU:

0 (nvbuf-mem-default): Default memory, cuda-device

1 (nvbuf-mem-cuda-pinned): Pinned/Host CUDA memory

2 (nvbuf-mem-cuda-device) Device CUDA memory

3 (nvbuf-mem-cuda-unified): Unified CUDA memory

For Jetson:

0 (nvbuf-mem-default): Default memory, surface array

1 (nvbuf-mem-cuda-pinned): Pinned/Host CUDA memory

2 (nvbuf-mem-cuda-device) Device CUDA memory

3 (nvbuf-mem-cuda-unified): Unified CUDA memory

4 (nvbuf-mem-surface-array): Surface array memory

enum GstNvVidConvBufMemoryType

src-crop

Pixel location: left:top:width:height of the input image which will be cropped and transformed into the output buffer. If the crop location is out of bound the values will be clamped to image boundaries of the input image. This property will be applied to all the input images in the batch

String

src-crop=”20:40:150:100”

dest-crop

Pixel location: left:top:width:height is the location in the output image where the input image will be transformed to. If the crop location is out of bound the values will be clamped to image boundaries of the output image. This property will be applied to all the output images in the batch

String

dest-crop=”20:40:150:100”

interpolation-method

Interpolation method:

0: Nearest

1: Bilinear

2: Algo-1 (GPU—Cubic, VIC—5 Tap)

3: Algo-2 (GPU—Super, VIC—10 Tap)

4: Algo-3 (GPU—LanzoS, VIC—Smart)

5: Algo-4 (GPU—Ignored, VIC—Nicest)

6: Default (GPU—Nearest, VIC—Nearest)

enum GstInterpolationMethod

interpolation-method=1 Default value is 6.

compute-hw

Type of computing hardware:

0: Default (GPU for gDPU, VIC for Jetson)

1: GPU

2: VIC

enum GstComputeHW

compute-hw=0 Default value is 0.

gpu-id

Device ID of GPU to use for format conversion

Integer, 0 to 4,294,967,295

gpu-id=0

output-buffers

Number of Output Buffers for the buffer pool

Unsigned integer, 1 to 4,294,967,295

output-buffers=4

bl-output

Flag applicable only for Jetson to output Block linear NVMM buffer for NV12 format. This is ignored for dGPU, Software memory (output buffer) and formats other than NV12

Boolean. Default: false

bl-output=1

flip-method

Type of rotation method. This flag is applicable for Jetson and GPU.

Flip method:

0: Identity (no rotation)

1: Rotate counter-clockwise 90 degrees

2: Rotate 180 degrees

3: Rotate clockwise 90 degrees

4: Rotate horizontally

5: Rotate across upper right/lower left diagonal

6: Rotate vertically

7: Rotate across upper left/lower right diagonal

enum GstNvDsVideoFlipMethod

flip-method=1 Default value is 0.

allow-odd-crop

Flag to allow the odd dimensions for source and destination crop rectangle

Boolean

allow-odd-crop=1 Default value is 1.

contiguous-buffers

Enable contiguous output buffers

Boolean

contiguous-buffers=1 Default value is 0.

disable-passthrough

Disable passthrough mode at init time

Boolean

disable-passthrough=1 Default value is 0.

copy-hw

Select hardware used for surface copies:

1: GPU

2: VIC

flags: readable, writable, controllable

enum GstNvCopyHWType

copy-hw=2 Default value is 1.

Property	Meaning	Type and Range	Example Notes
nvbuf-memory-type	Type of memory to be allocated. For dGPU: 0 (nvbuf-mem-default): Default memory, cuda-device 1 (nvbuf-mem-cuda-pinned): Pinned/Host CUDA memory 2 (nvbuf-mem-cuda-device) Device CUDA memory 3 (nvbuf-mem-cuda-unified): Unified CUDA memory For Jetson: 0 (nvbuf-mem-default): Default memory, surface array 1 (nvbuf-mem-cuda-pinned): Pinned/Host CUDA memory 2 (nvbuf-mem-cuda-device) Device CUDA memory 3 (nvbuf-mem-cuda-unified): Unified CUDA memory 4 (nvbuf-mem-surface-array): Surface array memory	enum GstNvVidConvBufMemoryType
src-crop	Pixel location: left:top:width:height of the input image which will be cropped and transformed into the output buffer. If the crop location is out of bound the values will be clamped to image boundaries of the input image. This property will be applied to all the input images in the batch	String	src-crop=”20:40:150:100”
dest-crop	Pixel location: left:top:width:height is the location in the output image where the input image will be transformed to. If the crop location is out of bound the values will be clamped to image boundaries of the output image. This property will be applied to all the output images in the batch	String	dest-crop=”20:40:150:100”
interpolation-method	Interpolation method: 0: Nearest 1: Bilinear 2: Algo-1 (GPU—Cubic, VIC—5 Tap) 3: Algo-2 (GPU—Super, VIC—10 Tap) 4: Algo-3 (GPU—LanzoS, VIC—Smart) 5: Algo-4 (GPU—Ignored, VIC—Nicest) 6: Default (GPU—Nearest, VIC—Nearest)	enum GstInterpolationMethod	interpolation-method=1 Default value is 6.
compute-hw	Type of computing hardware: 0: Default (GPU for gDPU, VIC for Jetson) 1: GPU 2: VIC	enum GstComputeHW	compute-hw=0 Default value is 0.
gpu-id	Device ID of GPU to use for format conversion	Integer, 0 to 4,294,967,295	gpu-id=0
output-buffers	Number of Output Buffers for the buffer pool	Unsigned integer, 1 to 4,294,967,295	output-buffers=4
bl-output	Flag applicable only for Jetson to output Block linear NVMM buffer for NV12 format. This is ignored for dGPU, Software memory (output buffer) and formats other than NV12	Boolean. Default: false	bl-output=1
flip-method	Type of rotation method. This flag is applicable for Jetson and GPU. Flip method: 0: Identity (no rotation) 1: Rotate counter-clockwise 90 degrees 2: Rotate 180 degrees 3: Rotate clockwise 90 degrees 4: Rotate horizontally 5: Rotate across upper right/lower left diagonal 6: Rotate vertically 7: Rotate across upper left/lower right diagonal	enum GstNvDsVideoFlipMethod	flip-method=1 Default value is 0.
allow-odd-crop	Flag to allow the odd dimensions for source and destination crop rectangle	Boolean	allow-odd-crop=1 Default value is 1.
contiguous-buffers	Enable contiguous output buffers	Boolean	contiguous-buffers=1 Default value is 0.
disable-passthrough	Disable passthrough mode at init time	Boolean	disable-passthrough=1 Default value is 0.
copy-hw	Select hardware used for surface copies: 1: GPU 2: VIC flags: readable, writable, controllable	enum GstNvCopyHWType	copy-hw=2 Default value is 1.