Running SSD mobilenet v2 object detection on saved file

I have checked cloned the jetson inference using

$ git clone -b python jetson-inference-python

in order to find the optimum threshold I would like to save the camera input as video (e.g. .mpg, or mp4, etc) and the then run the network (SSD mobilenet v2) for different threshold on the saved video sequence.

I found example which takes an image and not a video? Is there a sample code for saving camera input as a video file and then if there is a sample code to run object detection on the saved video file?

The only solution I thought is to save the input images as jpg then extend the to run on a number of images rather than one jpg, is this a good solution?


Hi am_merati, you could in theory modify the sample to read/decode compressed video file from GStreamer, but it may be easier for your testing to just convert the video to a sequence of JPG’s.

You could also try this SSD-Mobilenet sample if it is easier for you to modify: