Realsense Augmentor

Build error due to display problems in travis. 

End to end software to capture new objects using RGBD camera and augment them to get a annotated dataset to train deep nets

1. Pipeline to artificial generate annotated data for training deep learning models.
2. Pipeline includes starting from capturing images using provided camera (Realsense), generate semantic labels of the captured image and then generate the artificial images.
3. GUI required for the end user from capturing to labelling and generating new data.

Requirements

1. Ubuntu 16.04 (Testing for Ubuntu 18.04)
2. Intel Realsense Camera
3. Processer intel i5 or higher
4. python 3.5

Limitations

1. Number of classes captured should be more than or equal to 2.
2. Only one object in the scene to segment, not supported for multiple objects.

Installation

Build from source

Linux:

git clone https://github.com/santoshreddy254/easy_augment.git
cd easy_augment
./setup.sh

From pip

pip3 install easy-augment --user

Usage example

cd realsense_augmentor
python3 src/main.py

1. Start page will be as below and select the path to save the captured images.
2. On selecting capture next window will look like
3. On selectin Have Annotations Steps after selecting capture option and selecting save path
4. Next window will have image and mask of corresponding object. Capture as many as images per classe.
5. Click add to add new class label.
6. Click save to save the current displyed image and semantic label.
7. Click finish once done with capturing all the images.
8. Folder name \textbf{captured_data} in selected save path will have images, labels and labels.txt
9. Input parameters to generate artificial images need to be filled next window.
10. Click OK once setting up the parameters.
11. Folder name \textbf{augmented} in selected save path will have artificial images.

Steps after selecting Have annotations

12. Select the location of images and corresponding annotations

13. Select save path

14. Select labels.txt file path

15. Follow the steps from Step 9

Release History

• 1.0.2
  • Fixed bugs with augmeting data with labelme data.
• 1.0.1
  • Released on pypi.
• 1.0.0
  • First release for crowd testing.

Contributors

• Santosh Muthireddy – https://github.com/santoshreddy254
• Naresh Kumar Gurulingan - https://github.com/NareshGuru77
• Deepan Chakravarthi Padmanabhan - https://github.com/DeepanChakravarthiPadmanabhan
• M.Sc Deebul Nair - https://github.com/deebuls

Distributed under the MPL license. See LICENSE for more information.