Jetson TK1 Live Video Object Detection

I am working on a project to detect crop regions from a live video feed. I was looking forward to using tensorflow 1.0 object detection API on my Jetson TK1 to detect the weed regions. However, I couldn’t find a tutorial to install tensorflow 1.0. Is is possible to do the same on Jetson Tk1 or should I look for a different alternative like Caffe maybe? Thank you for your kind help.

Hi,

It’s recommended to use TX1 or TX2 for deep learning use case.

TK1 is 32-bit Ubuntu14.04 and only support cuDNN v2, can’t meet the requirement of TF-1.0.
Here is the installation script for Caffe on TK1:

Thanks.

Hi,
Thank you for clarifying my doubt. I tried to build caffe and managed to do so successfully, but seems like real-time object detection is tough on jetson tk1 using caffe.
I currently have used jetpack 3.1 and have l4t21.5 on my jetson. It comes with opencv 2.4.13. I wanted to know it’s it possible to use opencv 3.x on jetson along with gpu support. If you so can you point me to some resources for the same.
Thank you once again.

Hi,

It’s workable and here is a script to build OpenCV-3.4 on TX2 for your reference:
https://github.com/AastaNV/JEP/blob/master/script/install_opencv3.4.0.sh

Please remember to correct the GPU architecture for TK1 before executing.

cmake -D WITH_CUDA=ON -D CUDA_ARCH_BIN="<b>3.2</b>" ...

Thanks.