Low Execution time when Running 4 Cameras in parallel (Drivenet N-cameras sample)

Hi I want to improve the execution time for running 4 cameras in parallel (online with 30FPS cameras). The time for 1 camera is 6ms. I have observed that the sample uses only one RAW-RGB pipeline for processing images. Is this the reason behind the lower FPS. Can multi-threading and running 4 instances of the pipeline help?

Dear @a.a.menezes,
What does that 6ms include? Just conversion time? You want to run 4 conversion pipelines in parallel instead of 1 in batchmode?

Dear @a.a.menezes,
Any update can be provided on this issue?

Let me rephrase my question, when I run DriveNet Ncameras sample with 1,2,3 &4 simultaneous video inputs on my host machine, I observe that the time taken is 6.5,13,20,27ms respectively. Should not batch processing(DrivenetNcameras API) have the same execution time irrespective of camera count?

6ms includes only the image conversion processing(RAW->RGB). Is batch mode possible? If yes could you please provide an example?

Dear @a.a.menezes,
I could see in the sample that, RAW to RCB conversion using dwISP module happens on single pipeline. Hence, all those calls gets executed serially. You may check creating seperate pipelines(using different CUDA streams). Note that, the conversion calls gets executed in parallel only when enough resources are available. if not, they gets serialized even if we create seperate pipeline.