installation_help.txt

###Install Opencv3.4.11 
1. Follow this link and choose the second method (install from source) on Ubuntu 20.04
https://linuxize.com/post/how-to-install-opencv-on-ubuntu-20-04/
OR Ubuntu 18.04
https://linuxize.com/post/how-to-install-opencv-on-ubuntu-18-04/
2. For opencv 3.4.11, use “git checkout 3.4.11” inside opencv and opencv_contrib folders
3. Now, you can run opencv in python3
4. For opencv to work in C language, we need to install the missing library
“sudo apt-get install libopencv-highgui-dev”

### Uninstall Opencv
1. Inside the build directory of opencv (opencv_build/opencv/build), if installed from source
$ sudo make uninstall 
https://stackoverflow.com/questions/9276169/removing-all-installed-opencv-libs/#answer-50808950
2. Follow this
https://medium.com/@changrongko/opencv-how-to-uninstall-opencv-dfe1a5a50193
3. Uninstalling like this also removes opencv files from ROS, so also make sure to reinstall ros-melodic with instructions here
https://stackoverflow.com/questions/58484540/no-rule-to-make-target-libopencv-calib3d-so-3-2-0-but-opencv-3-4-1-installed-wh#_=_
Optional: if python 2 and 3 are showing different cv versions, you may delete the one in python2 by the same method but make sure to only delete “*opencv*3.2*” in the `find` command line

### install nlohmann
```$ sudo apt install nlohmann-json-dev```
- The above instructions will make it build successfully. After that just try to run ```$ python3 scripts/mask_rcnn_rospy.py``` and then install the pip packages as necessary.

###Apex installation, 
- follow this https://github.com/NVIDIA/apex but instead of 3rd line, run this
```$ sudo python3 setup.py install```

###<tf2_sensor_msgs, vision_msgs>
- go to catkin_ws/src and install them by their github link that can be found on ros wiki.
OR follow the recommended way:
- $ sudo apt install ros-melodic-package-name

### CPARSE
sudo apt-get install libsuitesparse-dev

### Install Eigen
1. Download the required version of eigen3
2. open the “install” file and follow the second method (which uses cmake)

### Install CUDA(easier way is through downloading the ZED SDK below)
On linux, follow
https://docs.nvidia.com/cuda/cuda-installation-guide-linux/
For installing samples, look for that step in a similar installation procedure
https://docs.nvidia.com/cuda/cuda-quick-start-guide/index.html
To run samples, we need to install third party libraries
https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#install-libraries

### Uninstall CUDA
sudo apt-get purge nvidia*
sudo apt-get autoremove
sudo apt-get autoclean
sudo rm -rf /usr/local/cuda*
sudo apt-get purge cuda*

### Install ZED SDK
- installed ZED SDK as shown on website with CUDA 11.1 automatically installed
https://www.stereolabs.com/docs/installation/linux/

- in order to use “nvcc” command, follow this link to export the toolkit to the environment PATH. https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#mandatory-post
Solution found here: 
https://stackoverflow.com/questions/53422407/different-cuda-versions-shown-by-nvcc-and-nvidia-smi#answer-53504578

### Install pytorch
https://www.codegrepper.com/code-examples/shell/how+to+install+pytorch+0.4.1


### (optional) Install Eclipse (recommended for CUDA)
https://websiteforstudents.com/how-to-install-eclipse-ide-on-ubuntu-20-04-18-04/
Install nsight Eclipse plugins for cuda
https://docs.nvidia.com/cuda/nsightee-plugins-install-guide/index.html

### (optional) Install dataset_metapackage to be able to run Jesse’s vdoslam_tests package
https://gitlab.acfr.usyd.edu.au/its/dataset_metapackage

### (Optional & not directly associated with this project) 
Running MASK-RCNN implementation

- Follow this tutorial:
https://www.analyticsvidhya.com/blog/2019/07/computer-vision-implementing-mask-r-cnn-image-segmentation/
- install pycocotools
- installing correct versions of keras and tensorflow
tensorflow==1.9.0
keras==2.2.0
- Working maskrcnn jupyter notebook example
https://github.com/matterport/Mask_RCNN/files/4499391/MaskRCNN_180420.zip

### (Optional & not directly associated with this project)
Running Flownet2 implementation

1. Follow this tut:
https://github.com/NVIDIA/flownet2-pytorch
2.  install the required packages using pip3
3.  install pytorch 1.4.0 by changing the version number this link:
https://www.codegrepper.com/code-examples/shell/how+to+install+pytorch+0.4.1
4. download the MPI Sintel dataset here. Must choose the complete dataset [see issue]
http://sintel.is.tue.mpg.de/downloads
5. More complete tutorial for running flownet2 [includes creating the model]
https://cloud.tencent.com/developer/article/1461933
Running inference:
$ python3 main.py --inference --model FlowNet2 --save_flow --inference_dataset MpiSintelClean --inference_dataset_root /home/tranks/Downloads/MPI-Sintel-complete/training  --resume /home/tranks/Downloads/FlowNet2_checkpoint.pth.tar 
OR
$ python3 main.py --inference --model FlowNet2 --save_flow --inference_dataset ImagesFromFolder --inference_dataset_root /home/tranks/Downloads/MPI-Sintel-testing/test/clean/wall/  --resume /home/tranks/Downloads/FlowNet2_checkpoint.pth.tar 

6. format of .flo files
https://vision.middlebury.edu/flow/code/flow-code/README.txt
7. running the .flo files (branch from step 5 since code is incopiable)
https://github.com/georgegach/flowiz/blob/master/demo/test.ipynb
https://github.com/georgegach/flowiz/blob/master/flowiz/flowiz.py
8. making a video out of images
https://stackoverflow.com/questions/44947505/how-to-make-a-movie-out-of-images-in-python

EXTRA notes:
    • to use alternative models such as Flownet2C, download the “checkpoints” folder for that model on the official github and use it instead.
    • Flownet2 can’t run on 2MB-size images (saying out of CUDA memory) but Flownet2C can.