System requirements¶
Contents
Base requirements¶
Before installing Pose-Trigger, make sure you have set up the following hardware:
A linux computer (we tested on Ubuntu 18.04 LTS)
a 16-bit monochrome video camera from ImagingSource (e.g. refer to the Reference setup specifications).
Note
Other Video4Linux2-compliant cameras should also work with a few adjustments in the code, but will require some efforts.
For a faster working of DeepLabCut, NVIDIA graphics board with a large amount of RAM is required.
Note
For example, running DeepLabCut on ResNet-50 requires ~10.6 GB of RAM, so we use GeForce RTX 2080 Ti that has 11 GB on-board RAM (refer to the Reference setup specifications).
Requirements for trigger-output generation¶
In addition to the pose-estimation feature, the trigger-output feature requires the followings:
The trigger-output server (“FastEventServer”).
An output board based on Arduino or its clone.
For Intel 64-bit CPUs, Pose-Trigger comes with the working FastEventServer program; you don’t need to install it manually. For other architectures (e.g. AMD and ARM CPUs), refer to Appendix: Compiling FastEventServer.
Preparation of the Arduino-based output board may be non-trivial. Please refer to Appendix: Preparing an Output Board.
Reference setup specifications¶
We develop and test Pose-Trigger in the following environment:
Hardware¶
Part name |
Model type |
|---|---|
CPU |
3.7 GHz Core i7-9700K |
RAM |
64 GB DDR4-3200 |
GPU |
NVIDIA GeForce RTX 2080 Ti (11 GB RAM) |
Camera |
ImagingSource DMK 37BUX287 |
Output board |
Arduino UNO, rev. 2 (clone), with arduino-fasteventtrigger |
Software¶
Software |
Specification |
|---|---|
Operating system |
Ubuntu 18.04 LTS |
Python environment |
Anaconda3, Python 3.7.7 |
CUDA Toolkit |
version 10.1 (through conda) |
Tensorflow |
version 1.13.1 (tensorflow-gpu package of conda) |
DeepLabCut |
version 2.1.3 |
NumPy |
version 1.19.1 (through conda) |