A small set of startup scripts to install everything we need to run image and video retargeting script on amazon EC2.

The main script is boostrap that installs:

• python 2.7 (instead of 2.6)
• numpy 1.9
• opencv 2.4.9 with ffmpeg support
• Dropbox
• py-video-retargeting repository
• py-seam-merging repository
• py-tvd repository

After lanching an Amazon EC2 instance:

ssh using the certificate given _ssh -i CERTIFICATE.pem ec2-user@IP_ADDRESS_
#change to amazon shell
sudo yum update
sudo yum install -y git
git clone https://github.com/PNProductions/cloud-python-opencv
./bootstrap

To check if everything is installed properly run

./test

You'll get a version of each package installed, for example:

Python version --> 2.7.5
Numpy version --> 1.9.1
Opencv version --> 2.4.9
Cython version --> Version: 0.21.1
Image seam merging version --> Version: 0.2.4
Video seam merging version --> Version: 0.1
TVD version --> Version: 1.0
Dropbox is running correctly

##Example usage The main script is main.py, that can launch retargeting script both on images and on videos.

To start the algorithm with the default parameters run:

python main.py

You can also use the following options:

• -s SEAM: specify the seam to use (default is 1 for videos and image_width/2 for images)
• -f FRAME: specify the frame number to consider (default is 10)
• -p PATH: specify the path of the input videos ot photos (specify a folder not a single file), default is testing_videos/downsample/japan/
• -i: save the importance map
• -nv: do not save motion vector
• -g: use motion vector for the whole video and not frame by frame
• -d: Launch with Dropbox support, only for ec2-server. Copy the result in the dropbox folder and remove the old result
• -m {seam_merging,seam_carving,time_merging}:
  • seam_merging_gc: use seam merging algorithm with graph cut (default)
  • seam_merging: use seam merging algorithm with dynamic programming (not available for videos)
  • seam_carving: use Rubinstein seam carving method with forward energy
  • time_merging: apply a temporally resize instead that a width resize. A seam is a frame, so, if you want to delete 10 frame from the video use the option -s 10. This method is available only for videos

To avoid borken pipe error and to get the script to ignore the hangup signal and keep running use: nohup python main.py