DRA750: The questions about GPU-less 3D surround view using the mesh warp/LDC HWA

user4684340

Part Number: DRA750

Hi,

In the following link, the paper mentions GPU-less 3D surround view using the mesh warp/LDC HWA. There is the statements "The mesh-based surround view supports unlimited predefined virtual view points.

The PC-based tool (provided by TI) creates mesh tables and virtual car overlay images for each virtual viewpoint and stores them in memory."

http://www.ti.com/lit/wp/spry308/spry308.pdf

So I want to know the size of one view point's LUT file. Because I need 200 view points, I wonder whether I can put all LUT files in the device's storage.

Mason Su

Best regards

over 6 years ago

0 Jesse Villarreal over 6 years ago

TI__Expert 5485 points

Mason,

Our expert is out of the office this week. I have forwarded this to him, but due to his travel, response may be delayed.

Jesse

0 Shashank Dabral over 6 years ago in reply to Jesse Villarreal

TI__Intellectual 1530 points

Hi Mason

The typical size for one view (including LUTs, car model and blend tables) is around 220 KB. So for 200 views we are looking at around 44 MB of storage. This would be in addition to code size etc. Overall a 64 MB flash is recommended for this configuration.

Regards

Shashank

0 user4684340 over 6 years ago in reply to Shashank Dabral

Intellectual 290 points

Hi Shashank,

What is the resolution of the blended image for this 220kb LUT file?

I need to create a LUT file for a 1280x720 yuv image. In my understanding, I need 8 bytes for each pixel, so the lut file size should be at least 1280x720x8(about 7MB).

Mason Su

Best regards

0 Shashank Dabral over 6 years ago in reply to user4684340

TI__Intellectual 1530 points

Hi Mason,
The data was for a 1008x756 display size assuming 4 camera inputs and blend table along with car model. The LDC mesh is subsampled and we use that in addition to compression (typically around 10x lossless) to hit the 220KByte range. The size scales linearly, so you can use the data to estimate the size for your desired configuration.

Regards
Shashank

0 user4684340 over 5 years ago in reply to Shashank Dabral

Intellectual 290 points

Hi Shashank,

I am implementing the similar GPU-less 3d surround view(use dsp) in TDA2X, so I wonder what compression method TDA3X uses?

Best regard,

Mason Su

0 Shashank Dabral over 5 years ago in reply to user4684340

TI__Intellectual 1530 points

We used the LZ4 compression decompression

A reference implementation can be found @

https://github.com/lz4/lz4

Shashank

0 user4684340 over 5 years ago in reply to Shashank Dabral

Intellectual 290 points

Hi Shashank,

I use lz4 to decompress 500kB file in TDA2x which costs 90ms. So I wonder whether I can speed up the decompression. And can you tell when TDA 3x demo decompress the lut files (eg: decompress all view points at the startup or decompress one view point when this view point is needed).

Best regards
Mason Su

0 Shashank Dabral over 5 years ago in reply to user4684340

TI__Intellectual 1530 points

Hi Mason,

The decompression occurs on power on and not on the fly. Also, to improve boot time we can either store the first view as uncompressed or just uncompress the first view and start the display. All other views are decompressed after the display turns on.

The decompression algorithm runs on the DSP and even though we haven't tried optimizing it, it should be possible to speed it up significantly.

Regards

Shashank

Processors

Processors forum

DRA750: The questions about GPU-less 3D surround view using the mesh warp/LDC HWA