SK-AM62A-LP: CSI 1920x1200 @120 FPS Resolution with AM62A34

Hello Team

Please provide your feedback on above query. 

Thanks

Hi Yogesh

As per datasheet we can see capabilities:

MJPEG Encoding capability = 416 Mpixels/Sec
ISP capability = 315 MPixels/Sec
CSI capability = 1.5 Gbps/lane = 6Gbps (On 4 lane)

So as per above information is seems 1920×1200×120fps(276MPixel) streaming should be supported without any performance impact.

Reason for asking this question here is to get advice if we are missing any corner case with respect to processor architecture then we can catch it and consider it in current design phase of product development. Let us know if we are missing anything here.

For MJPEG encoding if we consider 416 MPixels/Sec then it should support 1920×1200×120fps resolution but as per below screen shot it is mentioned the capability as 1080p@60. So can we consider it's maximum capacity of MJPEG encoding is not beyond 1080p@60?

Thanks

Hi Jaimin,

Also please see the latency for 1920x1080@30fps is about ~8ms

Please refer the below link especially the section 3.2.2.20.8 

https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-edgeai/AM62AX/09_00_00/exports/docs/linux/Foundational_Components/Kernel/Kernel_Drivers/E5010_JPEG_Encoder.html

Hope this helps.

Best Regards,

Suren

Hi Suren,
Thanks for your reply.

Above given latency information is true when we capture multiple images and derive average value for the same.

We have application to capture single image from sensor and with that following observation.

I have check latency by using this gstreamer command:

GST_DEBUG_FILE=/run/trace_jpeg.log GST_DEBUG="GST_TRACER:7" GST_TRACERS="latency(flags=pipeline+element+reported)" gst-launch-1.0 -v v4l2src device=/dev/video-rpi-cam0 io-mode=dmabuf-import num-buffers=1 ! video/x-bayer, width=1920, height=1080, framerate=30/1, format=bggr ! tiovxisp sink_0::device=/dev/v4l-rpi-subdev0 sensor-name="SENSOR_SONY_IMX219_RPI" dcc-isp-file=/opt/imaging/imx219/linear/dcc_viss.bin sink_0::dcc-2a-file=/opt/imaging/imx219/linear/dcc_2a_10b.bin format-msb=9 ! video/x-raw, format=NV12, width=1920, height=1080, framerate=30/1 ! v4l2jpegenc output-io-mode=dmabuf-import ! fakesink

In that I have captured 1 frame and check latency using below script:

/opt/edgeai-gst-apps/scripts/gst_tracers# ./parse_gst_tracers.py /run/trace_jpeg.log

Results: 

 
So observation is for single image capture latency is very high for all elements. Is there a way using we can achieve same lower latency when we capture and process single image?

Thanks 

Hello  Suren Porwar,

any update in this case ?

Thanks.

Jaimin,

Could you please share complete picture of what you are trying to achieve here.

Thanks

Yogesh

Yogesh,

As per TI tiovxisp taking 8ms latency.

that is true when we capture multiple images and derive average value for the same. but when I am trying to capture 1 frame and in that I am getting this latency.

 

So observation is for single image capture latency is very high for all elements. Is there a way using we can achieve same lower latency when we capture and process single image?

Jaimin,

I have assigned your query to our CSI/VPAC expert. Will also try to run the pipeline for single image capture and see if the behavior is same as you are observing.

Best Regards,

Suren

Yes, I measured performance with capturing multiple image and with that we are getting proper latency data.

Thank you for providing explanation. We can close this thread.