Install Annolid

Requirements

  • Ubuntu / macOS / Windows

  • Python >= 3.6

  • [PyQt4 / PyQt5]

Set up Anaconda virtual environment

To run Annolid, we suggest using the Anaconda package/environment manager for Python (free for personal use). Download and install the Anaconda environment first. Then do the following, using the bash shell in Linux or the conda command line (= Anaconda Prompt) in Windows:

# For MacOS or Linux
conda create --name=annolid python=3.7
source activate annolid
# conda install -c conda-forge pyside2
# conda install pyqt
# pip install pyqt5  # pyqt5 can be installed via pip on python3
# conda install git

Commented-out lines are optional; use them if necessary.

# For Windows
conda create --name=annolid python=3.7
conda activate annolid
conda install -c conda-forge pyside2
conda install git

This creates a virtual environment called “annolid” into which you will install Annolid and all of its dependencies, along with whatever other Python tools you may need. Python 3.7 is recommended, as it is the version being used for Annolid development.

Install ffmpeg

conda install -c conda-forge ffmpeg

Install Annolid

Clone the code repo from Github, change into the new directory, and install

git clone --recurse-submodules https://github.com/healthonrails/annolid.git
cd annolid 

# install the package (be sure to include the space and the period after the -e)
pip install -e .

Note: if you get the following error, please try to install ffmpeg and then install decord from source as described here.

ERROR: Could not find a version that satisfies the requirement decord>=0.4.0

To fix error, “Failed to load platform plugin “xcb” while launching qt5 app on Linux

sudo apt install --reinstall libxcb-xinerama0

Launch Annolid

Typing “annolid” will bring up the Annolid GUI, which is based on the excellent LabelMe package.

# be sure to activate the annolid virtual environment every time you restart Anaconda; the conda shell prompt should read "(annolid)"
source activate annolid   # on Linux/MacOS
conda activate annolid    # on Windows
# then start Annolid (a GUI window will appear)
annolid

# Optionally, you can use a predefined label.txt file as follows.
annolid --labels=/path/to/labels_custom.txt

Install Detectron2 locally

Note: if you intend to process your tagged videos using Google Colab (which you should do unless you are using a workstation with a higher-end GPU), then you do not need to install Detectron2 on your local machine, and you can ignore the following steps.

Requirements

  • Windows, Linux or macOS with Python ≥ 3.7

  • PyTorch ≥ 1.5 and torchvision that matches the PyTorch installation. Install them together at pytorch.org to make sure of this.
    Presently, the combination of torch 1.8 and torchvision 0.9.1 works well, along with pyyaml 5.3, as shown below.
    For purposes of using Annolid, it is OK to downgrade pyyaml from its current version to 5.3.

Install Detectron2 dependencies:

pip install pyyaml==5.3 
pip install pycocotools>=2.0.1
pip install torch==1.9.0+cu102 torchvision==0.10.0+cu102 -f https://download.pytorch.org/whl/torch_stable.html

Install Detectron2:

See https://detectron2.readthedocs.io/tutorials/install.html for further information.

import torch
assert torch.__version__.startswith("1.9")    
pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu102/torch1.9/index.html

### Install Detectron2 on Windows 10

`git clone https://github.com/facebookresearch/detectron2.git`
`cd detectron2`
`pip install -e .`

Note. If you encounter an error on windows with message says:

in _run_ninja_build raise RuntimeError(message) RuntimeError: Error compiling objects for extension Please go to the link https://support.microsoft.com/en-us/topic/the-latest-supported-visual-c-downloads-2647da03-1eea-4433-9aff-95f26a218cc0 and download x64: vc_redist.x64.exe. Please click and install it. After restart, you can cd to detectron2 folder and run the following command. pip install -e ..

Install Detectron2 on Google Colab

Instructions will be posted here presently. (Colab uses CUDA 10.2 + torch 1.9.0).
Open In Colab

Optional: Install older version of Pytorch for YOLACT

Note: YOLACT models are less accurate comparing to Mask-RCNN in Detectron2. However, it is faster in terms of inference.

# DCNv2 will not work if Pytorch is greater than 1.4.0
!pip install torchvision==0.5.0
!pip install torch==1.4.0

For more information, please check https://github.com/healthonrails/annolid/blob/master/docs/tutorials/Train_networks_tutorial_v1.0.1.ipynb and https://github.com/healthonrails/yolac.