The tivxImgPreProcNode can be used to convert YUV420 to planar and padded RGB used as input for TIDL. In PSDK 7, this has been speed optimized by using DMA, and is also used in several demos that use the processing:
camera capture > VISS -> LDC --(NV12)-> ImgPreProc -> TIDL -> ...

The tivxImgPreProcNode assumes the YUV420 input format to be BT601 with reduced range (which was apparently the format used in some early demos), while the OpenVX standard NV12 image has a color format of full range BT709.

In other words, the YUV to RGB color conversion performed by the tivxImgPreProcNode for all current TI vision_apps demos using the above input processing (LDC -(NV12)-> ImgPreProc) is incorrect, since the NV12 input images to the tivxImgPreProcNode have BT709 full range color format, and not BT601 reduced range

In addition, VISS and DSS both support BT709 full range coefficients, so tivxImgPreProcNode causes clipping in highlight and shadow region.

Please refer to below thread for more details.
https://e2eprivate.ti.com/hyundai_mobis/hyundai-mobis_-_ep_automotive/f/adas-support-qna/18361/410-18361-adas_vp-tivximgpreprocnode-incorrect-nv12-color-conversion