What is the memory format of the output of dwDataConditioner_prepareData function?

Hi, I am trying to understand sample_object_detector_tracker from DriveWorks dnn samples. TensorRT models requires inputs to be in NCHW order. Is dwDataConditioner_prepareData() function’s output in NCHW order?

Thanks in advance.

Dear @f.ugurcali,
The dwDataConditioner_prepareData() API performs operations like mean value substraction, pixel value scaling and does not change image memory format.

So how can we achieve the conversions like NHWC to NCHW. Also I am trying to use my trained model as shown in dnn/sample_object_detector_tracker. My tensorflow model trained in NHWC order, I converted my model files to uff then converted it to bin file using trt-optimizer. Question in my mind is that does trt-optimizer changes the memory order for my model (from NHWC to NCHW), or can I supply input in NHWC order as in tensorflow?

Thanks in advance @SivaRamaKrishnaNV.

Dear @mugurcal,
You need to convert your model from NHWC to NCHW before feeding(Please check sampleMaskRCNN sample in TRT as an example). trt-optimizer just optimizes your model and does not change NHCW -> NCHW. I would recommend any model to test using TensorRT (like as a TRT sample or using trtexec) before integrating into DW DNN framenwork.

Thank you for your reply. My mind is clear now.

Dear @mugurcal,
for the future, please use your business email ID while posting topic. thanks