Skip to content

Latest commit

 

History

History
386 lines (295 loc) · 15.6 KB

NVIDIA.rst

File metadata and controls

386 lines (295 loc) · 15.6 KB
Raw

NVIDIA Jetson driver

Installation instructions

Installation instructions are provided in the README of the driver repository: https://github.com/alliedvision/linux_nvidia_jetson

Changes

For supported boards and JetPack versions, see the README of each driver release.

Changes in driver version 35.2.1-5.1.0

As of this version, the driver can be used with Allied Vision's Vimba X SDK: https://www.alliedvision.com/en/products/software/vimba-x-sdk/.

Note for Auvidea carrier JNX30-PD: Please contact Auvidea for a carrier hardware revision with a bug fix for the issue: On one of the 2-lane ports, no camera can be detected (clock and data lanes are mixed up).

Changes in driver version 4.0.0

As of this version, the driver can be used with Allied Vision's Vimba SDK v6.1.

Changes in driver version 3.0.0

As of this version, the driver can be used with Allied Vision's Vimba SDK v6.0: https://www.alliedvision.com/en/products/vimba-sdk/

Changes in driver version 2.0.0

Exposure active signals

The Jetson driver now supports exposure active signals. Note that the camera requires firmware version 00.04.00.34658 or higher. For information about exposure active, see the Alvium MIPI CSI-2 Cameras Direct Register Access Controls Reference: https://www.alliedvision.com/en/support/technical-documentation/alvium-csi-2-documentation.html

Note especially that exposure active output line mode must be set to Off before output lines can be changed.

Min/max values for exposure and gain

Min/max values for exposure auto and auto gain are available. Note that the values are limited to 32 bits.

Support of two cameras

You can connect and operate two Alvium cameras with CSI-2 interface.

Restrictions

Cropping

Alvium CSI-2 cameras with firmware version 0.1.0 don't support cropping of the width.

Image widths not supported by NVIDIA

NVIDIA's video input has restrictions concerning the image width depending on the pixel format. If the camera sends an image width not divisible by 16, 32, or 64 (according to the current pixel format), there is a padding at the end of each line. Some tools such as qv4l2 don't support padding bytes, so that these images are displayed incorrectly.

Workaround: Enable cropping (see BasicV4L2 in the examples repository) and use a supported value.

Supported number of lanes

Currently, the driver supports using 4 lanes.

Special features, troubleshooting

The V4L2 node provides some features that are no camera features, but features of the NVIDIA video input unit (VI). It is not recommended to change these features because the stream might fail or image data gets corrupted. Rebooting Jetson TX2 resets these features.

  • frame_timeout
  • height_align
  • write_isp_format
  • sensor_modes
  • disable_frame_timeout
  • disable_stride_aligment
  • low_latency_mode

V4L2 compliance tests

Some V4L2 compliance tests fail because of additional functionalities, but camera usage is not affected.

Tests that fail:

  • test VIDIOC_QUERY_EXT_CTRL/QUERYMENU
  • test VIDIOC G/S_PARM
  • test Cropping
  • test Composing

Supported pixel formats

The following tables lists supported pixel formats on Jetson boards of the current driver. Previous driver versions don't support 10-bit and 12-bit RAW formats. On Jetson TX2 and Jetson Xavier, 10-bit and 12-bit RAW formats deviate from the V4L2 standard and custom formats used. Compared to the standard formats, the custom format contains "TX2" or "XAVIER". The custom 4CC formats contain "Y" for Xavier and "2" for TX2.

Platform Pixel format name
Standard V4L2_PIX_FMT_Y10 ('Y10')
TX2 V4L2_PIX_FMT_TX2_Y10 ('J2Y0')
Xavier V4L2_PIX_FMT_XAVIER_Y10 ('JXY0')

Compared to the standard formats, the bits on Jetson TX2 and Xavier are shifted to the left and there are one or two zero bits at the beginning. In the example below, there are two zero bits and only B9 to B0 contain productive data. The low-order bits are undefined and therefore listed as "X".

10-bit and 12-bit formats are padded to 16 bits per channel.

V4L2_PIX_FMT_Y10 bit order in a 16-bit word:

0 0 B9 B8 B7 B6 B5 B4 B3 B2 B1 B0 X X X X

The Byte order is always identical with the standard pixel format.

Standard formats and byte layouts are documented at:

https://www.kernel.org/doc/html/v4.15/media/uapi/v4l/pixfmt-rgb.html

https://www.kernel.org/doc/html/v4.15/media/uapi/v4l/yuv-formats.html

Nano

On Nano, all supported pixel formats comply with the V4L2 standards.

Monochrome

Depth FourCC Enumerator
8 GREY V4L2_PIX_FMT_GREY
10 Y10 V4L2_PIX_FMT_Y10
12 Y12 V4L2_PIX_FMT_Y12

Bayer

Pattern Depth FourCC Enumerator
RGRG, GBGB 8 RGGB V4L2_PIX_FMT_SRGGB8
10 RG10 V4L2_PIX_FMT_SRGGB10
12 RG12 V4L2_PIX_FMT_SRGGB12
GRGR, BGBG 8 GRBG V4L2_PIX_FMT_SGRBG8
10 BA10 V4L2_PIX_FMT_SGRBG10
12 BA12 V4L2_PIX_FMT_SGRBG12
GBGB, RGRG 8 GBRG V4L2_PIX_FMT_SGBRG8
10 GB10 V4L2_PIX_FMT_SGBRG10
12 GB12 V4L2_PIX_FMT_SGBRG12
BGBG, GRGR 8 BGGR V4L2_PIX_FMT_SBGGR8
10 BG10 V4L2_PIX_FMT_SBGGR10
12 BG12 V4L2_PIX_FMT_SBGGR12

RGB

RGB3 is unsupported. X bytes are set to 0 by NVIDIA's video input unit.

Layout Depth FourCC Enumerator Memory layout
Byte 0-1-2-3
XRGB 8 BX24 V4L2_PIX_FMT_XRGB32 B-G-R-X
BGRX 8 XR24 V4L2_PIX_FMT_XBGR32 X-R-G-B

YUV

The camera outputs this format as UYVY instead of VYUY.

Layout Depth FourCC Enumerator Memory layout
Byte 0-1-2-3
XRGB 8 VYUY V4L2_PIX_FMT_VYUY Cr0-Y0-Cb0-Y1

Xavier AGX and NX

Bit order of 10-bit and 12-bit custom formats on Xavier AGX and NX:

Depth Bit order (X is undefined)
10 0 B9 B8 B7 B6 B5 B4 B3 B2 B1 B0 X X X X X
12 0 B11 B10 B9 B8 B7 B6 B5 B4 B3 B2 B1 B0 X X X

Monochrome

Monochrome standard format:

Depth FourCC Enumerator
8 GREY V4L2_PIX_FMT_GREY

Monochrome custom formats:

Depth FourCC Enumerator
10 JXY0 V4L2_PIX_FMT_XAVIER_Y10
12 JXY2 V4L2_PIX_FMT_XAVIER_Y12

RGB

RGB3 is unsupported. X bytes are set to 0 by NVIDIA's video input unit.

Layout Depth FourCC Enumerator Memory layout
Byte 0-1-2-3
XRGB 8 BX24 V4L2_PIX_FMT_XRGB32 B-G-R-X
BGRX 8 XR24 V4L2_PIX_FMT_XBGR32 X-R-G-B

Bayer

8-bit standard formats:

Pattern Depth FourCC Enumerator
RGRG, GBGB 8 RGGB V4L2_PIX_FMT_SRGGB8
GRGR, BGBG 8 GRBG V4L2_PIX_FMT_SGRBG8
GBGB, RGRG 8 GBRG V4L2_PIX_FMT_SGBRG8
BGBG, GRGR 8 BGGR V4L2_PIX_FMT_SBGGR8

10-bit and 12-bit custom formats:

Pattern Depth FourCC Enumerator
RGRG, GBGB 10 JXR0 V4L2_PIX_FMT_XAVIER_SRGGB10
12 JXR2 V4L2_PIX_FMT_XAVIER_SRGGB12
GRGR, BGBG 10 JXA0 V4L2_PIX_FMT_XAVIER_SGRBG10
12 JXA2 V4L2_PIX_FMT_XAVIER_SGRBG12
GBGB, RGRG 10 JXG0 V4L2_PIX_FMT_XAVIER_SGBRG10
12 JXG2 V4L2_PIX_FMT_XAVIER_SGBRG12
BGBG, GRGR 10 JXB0 V4L2_PIX_FMT_XAVIER_SBGGR10
12 JXB2 V4L2_PIX_FMT_XAVIER_SBGGR12

TX2

Bit order of 10-bit and 12-bit custom formats on TX2:

Depth Bit order (X is undefined)
10 0 0 B9 B8 B7 B6 B5 B4 B3 B2 B1 B0 X X X X
12 0 0 B9 B8 B7 B6 B5 B4 B3 B2 B1 B0 X X X X

Monochrome

Monochrome standard format:

Depth FourCC Enumerator
8 GREY V4L2_PIX_FMT_GREY

Monochrome custom formats:

Depth FourCC Enumerator
10 J2Y0 V4L2_PIX_FMT_TX2_Y10
12 J2Y2 V4L2_PIX_FMT_TX2_Y12

RGB

RGB3 is unsupported. X bytes are set to 0 by NVIDIA's video input unit.

Layout Depth FourCC Enumerator Memory layout
Byte 0-1-2-3
XRGB 8 BX24 V4L2_PIX_FMT_XRGB32 B-G-R-X
BGRX 8 XR24 V4L2_PIX_FMT_XBGR32 X-R-G-B

Bayer

8-bit standard formats:

Pattern Depth FourCC Enumerator
RGRG, GBGB 8 RGGB V4L2_PIX_FMT_SRGGB8
GRGR, BGBG 8 GRBG V4L2_PIX_FMT_SGRBG8
GBGB, RGRG 8 GBRG V4L2_PIX_FMT_SGBRG8
BGBG, GRGR 8 BGGR V4L2_PIX_FMT_SBGGR8

10-bit and 12-bit custom formats:

Pattern Depth FourCC Enumerator
RGRG, GBGB 10 J2R0 V4L2_PIX_FMT_TX2_SRGGB10
12 J2R2 V4L2_PIX_FMT_TX2_SRGGB12
GRGR, BGBG 10 J2A0 V4L2_PIX_FMT_TX2_SGRBG10
12 J2A2 V4L2_PIX_FMT_TX2_SGRBG12
GBGB, RGRG 10 J2G0 V4L2_PIX_FMT_TX2_SGBRG10
12 J2G2 V4L2_PIX_FMT_TX2_SGBRG12
BGBG, GRGR 10 J2B0 V4L2_PIX_FMT_TX2_SBGGR10
12 J2B2 V4L2_PIX_FMT_TX2_SBGGR12