FRCNN on TX2 board

[abhigoku10]

@radeonwu Hi can i have few queries 1.Can i run faster rcnn using tensorflow / tensor rt on the tx2 board 2. should i use pycaffe-frcnn to run the following models 3. i need to run frcnn+resnet on tx2 board

Thanks in advance

[PiyalGeorge]

@abhigoku10 , Have you found answers for your questions, if yes please share

[abhigoku10]

@PiyalGeorge ya i am able to run the faster rcnn on tx2 board . use the updated version of tensorflow 1.9

[PiyalGeorge]

@abhigoku how much speed did you get? I mean the fps?

[abhigoku10]

@PiyalGeorge i got the fps mentioned on the tensorflow site , and on top of that i have done few optimisations so i achieved good results

[PiyalGeorge]

@abhigoku10 okay. can you tell me the steps to run this repo?

[abhigoku10]

@PiyalGeorge i followed the similar steps which we follow when we run it on the PC system is there an specific question u want to ask

[PiyalGeorge]

@abhigoku10 thanks, one more question- I tried, but still didn't get the real-time fps(30fps). i searched in so many repos, discussions, but none didn't get real-time-fps for Faster-rcnn with jetson-tx2 and tensorRT. So did you get real-time fps?

[abhigoku10]

@PiyalGeorge Faster rcnn detection i got 30 fps with default pre-trained model , i did not implement tensor rt functionality for faster rcnn i still in progress

[PiyalGeorge]

@abhigoku10 , No way, i think no one has got 30fps with Faster-RCNN. correct me if im wrong.

[abhigoku10]

@Piyal I have got the mentioned fps. Even in nividia forum ppl have got it. I shall check it and verify it. Give couple of days time

[PiyalGeorge]

okay @abhigoku10 Thanks

[abhigoku10]

Hi @PiyalGeorge i ran my model again , sorry there was typo mistake in my code , i did not get 30fps it was around 10fps without any optimization . Any improvements on your end ?

[PiyalGeorge]

@abhigoku10 , Nope same here. didn't get 30 fps.