Artificial Intelligence, far from GAFA
“Borom Sarret” or “The Wagoner” (1963) film by Ousmane Sembene’s in black and white, reconstruction of color and thanks to artificial intelligence
- Car that drives itself
- Facial Recognition
- Targeted advertising
- Prediction on marketplaces
But far from the tech giants, such as Google, Amazon, Facebook. Here is another use of AI, the color reconstruction of old black and white films. This is possible thanks to AI.
“The Wagoner” by Ousmane Sembene in black and white
Here are the first images in color film
How did this idea come to me ?
Simply because we are still selling technological prowess far from our realities. I asked myself the question that is, how to make AI a reality in Africa without succumbing to the temptation to use algorithms or servers of GAFA. And my idea came to pay tribute to Ousmane Sembene to give back the color to “The Wagoner”.
Ousmane Sembene was a precursor, a visionary who mastered the technique of his time to reach the African masses who are largely illiterate. He adapted the cinema to our needs and our African culture. Long before Ryan Coogler.
It is important the future generation of African talent tailors AI to our needs, while bringing color, that is to say, on a hopeful note.
Is it my invention ?
No. The black and white image colorization learning program that I have written is based on the research of Richard Zhang, Phillip Isola and Alexei A. (https://arxiv.org/pdf/1603.08511.pdf ). They decided to tackle the problem using a convolutional neural network, which “Imagine” what an input grayscale image would look like if it were colored.
How it works ?
Deep Learning involves a computer program that has the ability to learn with a complex network of interconnected artificial neurons, which is nurtured by a database of more than one million images (http://image-net.org/).
The program converts all images from the RGB color space to the LAB color space. Use the L-channel as input to the network and practice predicting AB channels.
- The L channel encodes only the intensity of the brightness.
- Channel A encodes green → red.
- And the B channel encodes the blue → yellow.
Then combines the input channel L with the predicted AB channels and reconvert the LAB image to RGB.
If you are looking for more details on the image colorization algorithm and the deep learning model, please consult the official publication of Zhang (http://richzhang.github.io/colorization/ )
Is it safe to proceed ?
At the sight of this above image generated by the program, the first SLR was to do a search for the logo of “The Media Library of the three worlds” on google. I do not believe in my eyes, I came across the same picture with the same colors.
With an image of good quality, the conversion process works at 99%. But with 1% of cases where the program gives inconsistent results.
Our color reconstruction of “The Wagoner” is based on a YouTube video that is of average quality, which increases the number of inconsistencies.
With better image quality film we would be able to make a faithful replenishment.
When will the full version of the film?
The colored images, the work is not finished yet a calibration of the film is necessary for the respect of the work.
This work is done image by image, with a video of 18 minutes, we counted 27696 images to be treated, which made my motivation lower.
You are impatient to see all the film in color, encourage me by your claps, messages, shares or if not come to give me a hand.
What are the prospects for such work?
Many photos and videos of archives on the history of Senegal, which are at home, in our museums, libraries, national archives, Radio Television Senegal (RTS) and in many countries. Their color renditions can tell us more about our history.
Mamadou Diagne, The Digital Wagoner