Hello,
I am developing a project which is supposed to detect features from 2 video streams (2048x1536) and 50FPS each, in realtime. I have jetson nano but the FPS drops when I read two streams. Please suggest if Jetson Tx2 would work. I am using fast feature detection algorithm from OpenCV. I have no restrictions on space and budget. Also, since I am reading frame by frame in OpenCV would jetson give me required throughput i-e 50FPS?
Hi,
We would recommend try DeepStrem SDK on Jetson platforms. The latest version is 4.0.2:
https://devtalk.nvidia.com/default/topic/1068639/deepstream-sdk/announcing-deepstream-sdk-4-0-2/
The optimal solution on Jetson platforms is to use NVMM buffers and we developed DS SDK for deep learning inference usecases. The sample config file for Nano is source8_1080p_dec_infer-resnet_tracker_tiled_display_fp16_nano.txt, and you may install DS DSK via sdkmanager and try to run this config.
We don’t try fast feature detection from OpenCV on TX2 or Xavier. May see if some other users can share experience. Since OpenCV uses CPU buffers, higher CPU loading is expected than running DS SDK. CPU loading can be profiled in tegrastats