tkDNN
tkDNN copied to clipboard
centernet3d and centertrack3d, too many inferences results showing!
i have converted the cnet3d and centertrack 3d models to *.rt format models, however, when i ran the demos of demotrack and demo3D, i got lots of detections per frame for both models, i don't know why this happened, my environment is rtx3090 nvidia card, installed tensorrt8.2GA and cudnn8.X versions, please give me some advices to solve this weired issues!! thanks a lot, here are results showing :
i noticed that this repo hasn't tried to run on RTX3090 tensorrt8.2 yet,, does anyone have silimar issues with mine?
Hi ,I just tested ctrack and cnet3d 3d demos with tensorrt 8.4 and they seem to be working fine for me on linux.They should work with tensorrt 8.2 also.
@perseusdg , it is so weird, it truely looked like lots of detection results just showing as i detached above, i have tried to deploy the same program i compiled at jetson xvaier, it looked good , just like official demo showing. confused !
I can't be exactly sure of what the issue might be ,but i can speculate that it is probably a windows thing , i sometimes get a segmentation fault when running ctrack and cnet3d on windows,.