![conda install opencv 3.4.1 conda install opencv 3.4.1](https://i.stack.imgur.com/cNrbG.jpg)
- #CONDA INSTALL OPENCV 3.4.1 HOW TO#
- #CONDA INSTALL OPENCV 3.4.1 UPGRADE#
- #CONDA INSTALL OPENCV 3.4.1 SOFTWARE#
But I’d like to highlight a few things first: Installing opencv-3.4.6 on Jetson Nano using my script is straightforward. I’d strongly suggest you to set up a swap file on the Jetson Nano DevKit since its memory is quite limited. Please go through the steps I described in Setting up Jetson Nano: The Basics. In case floating-point computation precision matters to your application, you should consider removing those FAST_MATH definitions in the cmake command. I care more about fast computation or higher frame rate (FPS). That is, I don’t care if opencv functions return images with pixel values offset by a few decimal points from the correct values. With this, I’m actually trading floating-point computation precision for speed ( reference). I configured ( cmake) my opencv-3.4.6 with -D ENABLE_FAST_MATH=ON -D CUDA_FAST_MATH=ON in the script.
#CONDA INSTALL OPENCV 3.4.1 UPGRADE#
If I build opencv-3.4.6 with dependencies on the older version (3.0.0) of protobuf in Ubuntu 18.04, I’d likely run into trouble when trying to upgrade protobuf libraries later on. TF-TRT (TensorFlow) might require a newer version of protobuf to work well.Double free issue of protobuf (due to protobuf code in opencv) would cause caffe to crash.So I choose to use Qt (with OpenGL) backend instead of the default GTK+ backend. I don’t want my opencv-3.4.6 build to have any dependencies on ‘protobuf’ for the following reasons.
![conda install opencv 3.4.1 conda install opencv 3.4.1](https://cdn-ak.f.st-hatena.com/images/fotolife/o/ossyaritoori/20170524/20170524220154.png)
![conda install opencv 3.4.1 conda install opencv 3.4.1](https://media.springernature.com/original/springer-static/image/chp%3A10.1007%2F978-1-4842-6431-7_3/MediaObjects/502073_1_En_3_Fig30_HTML.jpg)
In my script I use opencv-3.4.6 since that’s the latest 3.4.x release as of the time of this writing. I actually purge the pre-installed opencv-3.3.1 packages at the beginning of my istall_opencv-3.4.6.sh script.Īlthough NVIDIA provides a script for installing opencv-4.0.0 on Jetson Nano, I’m sticking with 3.4.x because Caffe does not build with opencv-4.x. That’s the reason why I need to build and install OpenCV by myself. The pre-installed opencv-3.3.1 on Jetson Nano does not support gstreamer functionalities (cannot utilize hardware H.264/H.265 codec on Jetson Nano). I tried to explain my considerations about picking the version of OpenCV and its configurations below. So be sure to check that post out if you’d like to know more details about my OpenCV configurations.
#CONDA INSTALL OPENCV 3.4.1 HOW TO#
I don’t want to repeat all information I’ve written in this previous post: How to Install OpenCV (3.4.0) on Jetson TX2. Then I’d show how I use the script to build, install and test OpenCV on my Jetson Nano. In this post, I’d explain why I choose a certain version and configuration of OpenCV. I aggregate all steps of building/installing OpenCV into a shell scripts, so that it could be done very conveniently.
#CONDA INSTALL OPENCV 3.4.1 SOFTWARE#
As a follow-up on Setting up Jetson Nano: The Basics, my next step of setting up Jetson Nano’s software development environment is to build and install OpenCV.