Thanks to AI, The Night Watch is complete after 300 years.
“You see that Rembrandt is even more genius than he already was” — Museadirector Taco Dibbits
For the first time since 1715, The Night Watch is shown in the way that Rembrandt intended.
The Night Watch is a portrait from 1642 by Rembrandt van Rijn. This portrait is one of the famous Dutch Golden Age Paintings.
The Night Watch is famous for its size(363 cm x 437cm or 11.91 ft × 14.34 ft) and the combination of light and shadow, which gives a dramatic effect. The Portrait was Rembrandt’s contribution to a project that included seven works, and six militiamen’s pieces.
In 1715 The Night Watch was being relocated, and during the process, they find out that the Painting was too large to fit in the new place. They have cut pieces off, and these pieces never have been found. The exact size of the original portrait and the amount that has been removed are unknown. The height was between 358,7cm(11.7 ft) and 420 cm (13.77 ft), and the width was between 479 cm (15.72 ft) and 523 cm (17.16 ft).
The Operation of The Night Watch
“I designed a sequence of three artificial neural networks to make a computer reconstruction of the missing parts of the Night Watch from a small copy that was made before it was cut down.” — Rob Erdmann
Gerrit Ludens painted the smaller copy of the Night Watch. This copy is used to recreate the missing pieces of the Night Watch.
A neural network learns by analyzing a large number of good samples. The moment when the neural network is trained to solve the samples. Then the neural network is prepared for solving similar problems.
For the Night Watch, it was accomplished in several steps. The first Neural Network marked the crucial points in the Night Watch and Ludens, such as the faces and the bodies. There were still differences between the Night Watch and Ludens. Some parts of the painting need to shift so that all the crucial points are equivalent. In this part, all of the figures in these paintings look similar. To complete the reconstruction of the Night Walk. One neural network needs to be trained for the color, and the other one must be trained for the style. The paintings are cutting into dozens of small pieces and offered as problems and solutions. The problem is always the copy of Ludens, and the solution is the night watch. The neural Networks repaint Ludens's version in the style of the Night Watch.
As you can see above, there are two new faces present on its left, where there is also a small child.
New Technology is used to the Night Walk Alive
The latest imaging technology is used with the latest computer technology. They have collected data of 51 terabytes from all the different imaging techniques. The list below shows the number of captures generated and the total file size per technique. Due to this technology, the researchers on the operation gained deeper insight into the painting than ever.
- Daylight photography (20 µm): 528 captures, 317GB
- Daylight photography (5 µm): 8.439 captures, 5.063GB
- UV photography (20 µm): 528 captures, 317GB
- Macro X-ray fluorescence (500 µm & 250 µm): 56 captures, 1.008GB
- Reflectance imaging spectroscopy -VNIR (168 µm): 80 captures, 355GB
- Reflectance imaging spectroscopy -SWIR (168 µm): 127 captures, 1.130GB
- Optical coherence tomography: 66 captures, 1.500GB
- 3D scanning — Artec Spider (3D): 127 captures, 886GB
- 3D scanning (15 µm):135.346 captures, 40.604GB
Total size: 51.181GB
The whole operation of the Night Walk revealed so much more information about the Night Walk or about Rembrandt. The researchers discovered hundreds of hundreds of microscopic ‘pimples’. These are called lead soaps break through the paint surface and cause minuscule paint loss. In oil painting is lead soap formation a well-known phenomenon, it usually occurs when a lead-based primer has been used. But for The Night Watch, Rembrandt used a so-called quartz ground.
The restored painting is open for visitors to view at the Rijksmuseum in Amsterdam in the upcoming months. The painting gives visitors a new experience and through artificial intelligence, we are able to witness this masterpiece of piece Rembrandt.
I recently wrote a blog in line with this topic and I hosted a virtual event about: Can Artificial Intelligence create better art than humans?
I would recommend to read to blog and join the meetup group to be notified of the upcoming virtual events:
Can Artificial Intelligence create better art than humans?
Let’s begin with what a generative adversarial network is. A Generative Adversarial Network is a machine learning…
DataML Nederland (Amsterdam, Nederland)
Welcome to our meetup DataML We organize events within Artificial intelligence, Data Science, and Business…