Add FCN8sDepthPredictionConcatFirst model to fcn_depth_prediction.py

[YutoUchimi]

Add a neural network implement of my paper in JSAI2019 to fcn_depth_prediction.py

• Add sample dataset of mirror
• Add FCN8sDepthPredictionConcatFirst network model
• Add training script
• Add doc for the training script

[k-okada]

failing due to missing imgaug

00:20:47.160 imageio requires Python '>=3.5' but the running Python is 2.7.17

00:21:39.111 ERROR: the following rosdeps failed to install
00:21:39.111   pip: command [sudo -H pip2 install -U imgaug] failed
00:21:39.111   pip: Failed to detect successful installation of [imgaug]

00:34:37.602 
                                                                                
  File "/workspace/ros/ws_jsk_recognition/src/jsk_recognition/jsk_recognition_utils/python/jsk_recognition_utils/datasets/depth_prediction.py", line 8, in <module>
00:34:37.602 
                                                                                
    import imgaug.augmenters as iaa
00:34:37.602 
                                                                                
ImportError: No module named imgaug.augmenters
00:34:37.602 

because of imageio. need to use system imageio , but we still need to use pip for xenial

python-imageio:
  debian:
    '*': [python-imageio]
    jessie:
      pip:
        packages: [imageio]
    stretch:
      pip:
        packages: [imageio]
    wheezy:
      pip:
        packages: [imageio]
  fedora:
    pip:
      packages: [imageio]
  gentoo: [dev-python/imageio]
  ubuntu:
    '*': [python-imageio]
    trusty:
      pip:
        packages: [imageio]
    xenial:
      pip:
        packages: [imageio]

[knorth55]

because of imageio. need to use system imageio , but we still need to use pip for xenial

@k-okada @YutoUchimi how can i fix this issue?

[YutoUchimi]

I rewrote data augmentation part so as not to depend on imgaug any more.