Tool/software:
Platform Details:
Processor: TDA4VM (Custom Board)
Linux SDK:
ti-processor-sdk-linux-j7-evm-08_01_00_07RTOS SDK:
pipeline Details:ti-processor-sdk-rtos-j721e-evm-08_01_00_13
We are working with a TDA4VM-based custom board and implementing a multi-camera application using a producer-consumer architecture.
Our goal is to optimize performance by subsampling the image resolution before writing it to shared memory. However, we still require fisheye correction using the LDC node, which previously worked when using full-resolution data.Working Pipeline (with LDC and full resolution):
capture --> VISS (1936x1220) --> Shared Memory (1936 * 1220 Producer) --> Shared Memory (1936 * 1220 Consumer) --> LDC (1920x1080) --> Scaler (480x320) --> Mosaic
This pipeline works correctly and applies fisheye correction using the default TI SDK LUT (for input: 1936x1220, output: 1920x1080).Modified Pipeline (subsampled, without LDC):
capture --> VISS (1936x1220) --> Shared Memory (960 * 608 Producer) --> Shared Memory (960 * 608 Consumer) --> Scaler (480x320) --> Mosaic.
We perform 2x subsampling by dropping alternate pixels before writing to shared memory. This pipeline also works, but as expected, the output exhibits fisheye distortion, since no LDC correction is applied. please find the output here:
Target Pipeline (subsampled with LDC):
capture --> VISS (1936x1220) --> Shared Memory (960 * 608 Producer) --> Shared Memory (960 * 608 Consumer) --> LDC (1920x1080 or 960*608) --> Scaler (480x320) --> Mosaic.
In this version, we want to maintain the performance benefits of subsampling, but apply fisheye correction via LDC.
Questions:Is it necessary to regenerate a new LDC LUT specifically for the subsampled resolution (960x608)?
What are the recommended tools or workflows to generate a custom LDC LUT for a new resolution such as 960x608 input → 960x608 output or 960x608 input → 1920x1080 output?
We would appreciate your guidance on how to resolve this issue and ensure correct distortion correction at the subsampled resolution.
Thank you.
