ImageDetectionExtraction not ticking with sufficient input on isaac 2020.2

ImageDetectionExtraction does not seem to tick when provided with input on isaac 2020.2 on x86 machine. The codelet has a time associated with it when viewing statistic (ex. Count: 0, median: 0.06, 90%: 0.07, max: 0.15), compared to a codelet that isnt running would normally have (ex. Count: 0, median: 0.0, 90%: 0.0, max: -inf). Cuda is working properly (v10.2 running on a 2080super), tested on an trt inference already within isaac 2020.2, not sure if this codelet even uses cuda either. Wondering if this is known, or there is an updated configuration compared to that of the object_pose_estimation package. Same issue when using image from v4l2 source. Tried providing “tick_period”: “20Hz” in the config with no results either.

Test app is attached below. Replace “apps/eclipseworkspace/facial/faces.jpg” with any image over 100x100 pixels, or if the image is smaller just modify the bounding box size to fit the image, as well as the down sample.

test.app.json (2.9 KB)

The ImageDetectionExtraction codelet does use CUDA and only ticks with synchronized inputs on the image and detections endpoints. In your graph, is the publisher for detections actually publishing synchronized timestamps with the image publisher?

Sorry for the super late response. They were not synced. After publishing on synchronized timestamps it works as expected. Thank you!