search

naisy / realtime_object_detection

Plug and Play Real-Time Object Detection App with Tensorflow and OpenCV. No Bugs No Worries. Enjoy!
MIT License
101 stars 36 forks source link

Faster RCNN on Jetson TX2 #51

Open sivagnanamn opened 5 years ago

sivagnanamn commented 5 years ago

Hi @naisy ,

Thanks a lot for sharing your work.

In the readme, you have mentioned Faster R-CNN: PC only

Were you able to get FasterRCNN+Inception (or) FasterRCNN+ResNet working on TX2 with TensorFlow? Is there any known issue doing so?

naisy commented 5 years ago

Hi @sivagnanamn,

TX2 has small memory capacity, therefore error occurs at run time. You can download from https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md

sivagnanamn commented 5 years ago

Thank you @naisy .

FasterRCNN+Inception (or) FasterRCNN+ResNet seems to throw cuda launch out of resources error while running on TX2. Especially in the places where tf.where operation is involved.

However, moving tf.where to CPU results in wrong detection. Have you come across this behaviour before?

naisy commented 5 years ago

Hi @sivagnanamn,

I have not confirmed the existence of detection error caused by processing tf.where on CPU.

The CUDA of ARM architecture seems to require more memory than X86 architecture. I hope that memory problem will be solved with Xavier and CUDA10.