DeepRes

Uses Deep learning and sub-pixel convolutional nets to create High(er) resolution videos from low(er) resolution videos ![alt text](Compare 4.png "") The Top frame is the input/original, the bottom frame is the output frame.

How To run (Tested on MacOS Sierra CPU)

Make Sure To Edit the MainScript.py for your settings

After Editing the MainScript.py

python3.5 MainScript.py

How to include audio(Not Tested)

Normally the audio is not added as a feature, but it can be done easily.Although if you want audio, the framerate of the video must match the framerate varible in the MainScript.py. To get the fps: First install moviepy(http://zulko.github.io/moviepy/) Then in MainScript.py

import os
from moviepy.editor import *
EnhancePyDir = " /Users/USERNAME/Downloads/DeepRes/enhance.py"
videoDirectory = "/Users/USERNAME/Downloads/ATEST.mp4"
OutPutFileDirBefore = "/Users/aneeshmuppidi/out1"
OutPutFileDirAfter = "/Users/aneeshmuppidi/out2"
clip = VideoFileClip(videoDirectory)
ClipFPS = clip.fps
FrameRate =  str(ClipFPS)
AudioOutputFileName = "/Users/USERNAME/audio" 
Filename = "filename"
Dir = " ATEST/"
DirectoryToSave = Dir + Filename + "%03d.jpg"
DirectoryToSaveX = Dir +Filename + "%03d_ne2x.png"
command = "ffmpeg -i " + videoDirectory + " -vn -acodec copy " + AudioFileName +".aac"
os.system(command)
command1 = "ffmpeg -i " + videoDirectory + " -r " + FrameRate + DirectoryToSave
os.system(command1)

command3 = "python3.5" + EnhancePYDir + " --zoom=2" + Dir + "*.jpg"
os.system(command3)
command4 = "avconv -framerate 15 -f image2 -i " + DirectoryToSave + "  -b 65536k " + OutPutFileDirBefore + ".mov"
os.system(command4)
command5 = "avconv -framerate 15 -f image2 -i " + DirectoryToSaveX +  " -b 65536k " + OutPutFileDirAfter + ".mov"
os.system(command5)
command6 = "ffmpeg -i " + videoDirectory + " -i " + AudioOutputFileName + " \ "
os.system(command6)
command7 = "-c:v copy -c:a aac -strict experimental " + OutPutFileDirAfter + "WithAudio.mp4 "

Requirements python3.5(MUST)

colorama

pillow>=3.2.0

Theano==0.8.2

Lasagne  (https://github.com/Lasagne/Lasagne.git@61b1ad1#egg=Lasagne==0.2-dev)

Homebrew

Scipy

How to install FFMpeg and Libav

brew install libav

brew install ffmpeg

The Model upscales the Video by 2x at the exact same resolution!

Examples Note: The FPS of the videos were purposely decreased drastically, because the time that the model was being built had a time limit (hackathon) the higher FPS the more quality the results will be. :

One frame Comparison :

One frame example(After 10 epochs) :

The orignal Video was Higher Res Than the input, but I converted to a low res video, for the input of the Model.... So you can see the model's power converting very low res videos to High-Res

Before Super Resolution Model : https://www.youtube.com/watch?v=-H7Oba1H00s

After Super Resolution Model : https://www.youtube.com/watch?v=v5yBOTp5M9M

Aneeshers/DeepRes