PowerFilm.AI

Inspiration

4K era is coming with an expected astounding capital annual growth rate of 25% for the next 5 years and the entertaining industry is experiencing a lucrative revenue of more than 3 trillion dollars. However, 4K televisions are not supported by sufficient 4K videos. This is because a lot of people do not have professional equipment to produce a high-quality 4K video or image. At the same time, a lot of YouTubers/ Vloggers are very popular on social media. They also need tools to produce entertaining and high-resolution productions. Our product, Powerfilm.ai is a premier product that goes in sync with the current technology development. There is just a huge potential with our Powerfilm.ai as both an entertainment and a professional editing software.

What it does

drawing

Powerfilm.ai has three main functions: audio style transfer, video super resolution, and image super resolution. With a single webpage, you have the abilities to make perfect videos and images that align with the current 4K technology and transform your voices with ease. This can greatly assist the general public in multiple ways that others cannot imagine before:

Produce perfect video and photo with 4K resolutions from less-than-ideal qualities
Make attractive videos with unprofessional equipment and attract much more viewers on different personal platforms such as YouTube and Blogs
Add surprising features to your own voices for entertaining purposes
Provide production features for professions, protective features for high-profile personnel, and entertaining features within friends and family members

Powerfilm.ai has a lot of advantages such as:

Big money saver for people without professional equipment
Boost revenue for bloggers and YouTubers in assisting them with perfect videos
Fully utilize the 4K technology with video improvement services
Much more fun provided for general public

Features

Image & Video Super Resolution

Help to enhance the resolution of an imaging system to produce a much clearer image and video in a short period of time and gives people more power to produce make their best videos.

Audio Style Transformation

Gives people to transform original audios to audios of other styles, which can be used in different areas such as entertainment features for communications between friends, personal privacy protection, and even for after-effects edition in a professional video editing service.

Accomplishments that we're proud of

We completed with a working prototype with all features available within time limit. We picked up a lot of skills such as front-end, back-end and machine learning.

How we built

We used Python Flask to rapidly build a simple server which renders a home page. This home page is built with JQuery and Bootstrap and interacts with the server via AJAX. The server is integrated with machine learning models (RDN, SRGAN and AutoVC) to handle tasks ( Image SR, Video SR and audio style transfer). The machine learning models are based on TensorFlow, PyTorch and Keras.

Difficulty faced

We met some issues in file format convertion and we managed to fixed them. In addition, we also do not have a GPU to train our model. Instead, we looked for pre-trained weights online and load them into the machine learning models for the features to work.

Technology used

Python Library

Alertify
JQuery
Bootstrap
Fontawesome
Flask
PyTorch
TensorFlow
Keras
ISR
wavenet_vocoder
sk-video
FFMPEG
Librosa
scikit-image
Numpy

Deep Learning Model

Residual Dense Net -- Image Super Resolution
SRGAN -- Video Super Resolution
Auto-VC -- Audio Style Transfer

Future Works

To further improve the speed of processing develop a more efficient processing model with AI and ML
Enable more features to be applied to the audio in the future