In this post, we compare portraits generated by a neural style transfer algorithm with VGG-19 CNN network and StyleGAN2 trained on FFHQ (Flickr-Faces-HQ). The input for the first algorithm is a minimal sketch, while the input for the second one is the output portrait of the previous system.
| Minimal sketch | → | Neural style transfer with VGG-19 CNN | → | Generative adversarial network StyleGAN2 |
|---|
There is no question that machines can appear to be and are creative. They can produce works of art, alone or in collaboration with humans. They can even judge and be aware of the quality of their artworks. But there is one crucial element of the creative process they don't possess, that of intent. So how about a new hybrid kind of creativity, where humans and machine are collaborating into creating artworks; the humans provide the intention and the initiative, and the machines their interpretation and vision.
This project investigates this new area of hybrid human-machine creativity. The human draws some simple sketches of faces on an iPad. Then, she provides a photograph of a person, depicting emotions. These two are given to a machine as an input to transform those sketches according to its own "qualia" ("how it feels like", intrinsic instances of experience) using a neural style transfer algorithm based on a pre-trained convolutional network. The results are surrealistic artworks, dark and dystopian at times, but with evident anthropomorphic features, where somehow the machine manages to enhance the feeling of simple drawings.
Neural style transfer has been widely used to transfer artist painting style to photographs creating new pieces of art. The opposite direction has motivated this project, and the latter investigates the ability to create photorealistic images. Can simple sketches be brought to life? Using only a minimalistic sketch, a photograph of a person and a neural style transfer algorithm, the machine can bring simple illustrations into life transferring the human characteristics, for example, the eyes, ears, mouth. The neural style transfer algorithm used in this work is trained on the pre-trained VGG-19 convolutional neural network and based on Leon A. Gatys' paper, "A Neural Algorithm of Artistic Style". "Synesthesia." is the outcome of this algorithm given a digital drawing and a person's photo.
“Synesthesia”. Deep learning art, 2017.
“Synesthesia”. Deep learning art, 2017.
Neural style transfer has been widely used to transfer artist painting style to photographs creating new pieces of art. The opposite direction has motivated this project, and the latter investigates the ability to create photorealistic images. Can simple sketches be brought to life? Using only a minimalistic sketch, a photograph of a person and a neural style transfer algorithm, the machine can bring simple illustrations into life transferring the human characteristics, for example, the eyes, ears, mouth. The neural style transfer algorithm used in this work is trained on the pre-trained VGG-19 convolutional neural network and based on Leon A. Gatys' paper, "A Neural Algorithm of Artistic Style". "Hidden sorrow." is the outcome of this algorithm given a digital drawing and a person's photo.
“On the eve of morrow, shadows and sorrow, hidden in inaccessible depths, one million internal deaths” ~ ZR
“Hidden sorrow.” Deep learning art, 2020.
Content basis for “Hidden sorrow.”
Neural style transfer has been widely used to transfer artist painting style to photographs creating new pieces of art. The opposite direction has motivated this project, and the latter investigates the ability to create photorealistic images. Can simple sketches be brought to life? Using only a minimalistic sketch, a photograph of a person and a neural style transfer algorithm, the machine can bring simple illustrations into life transferring the human characteristics, for example, the eyes, ears, mouth. The neural style transfer algorithm used in this work is trained on the pre-trained VGG-19 convolutional neural network and based on Leon A. Gatys' paper, "A Neural Algorithm of Artistic Style". "Morphe." is the outcome of this algorithm given a digital drawing and a person's photo.
“Morphe.” Deep learning art, 2017.
Content basis for “Morphe.”
Neural style transfer has been widely used to transfer artist painting style to photographs creating new pieces of art. The opposite direction has motivated this project, and the latter investigates the ability to create photorealistic images. Can simple sketches be brought to life? Using only a minimalistic sketch, a photograph of a person and a neural style transfer algorithm, the machine can bring simple illustrations into life transferring the human characteristics, for example, the eyes, ears, mouth. The neural style transfer algorithm used in this work is trained on the pre-trained VGG-19 convolutional neural network and based on Leon A. Gatys' paper, "A Neural Algorithm of Artistic Style". "Ode to identity." is the outcome of this algorithm given a digital drawing and a person's photo.
Ode to Identity celebrates the need for unique personal identities, finding who you are and your real voice. Identifying oneself in a group is a human longing and need. Everyone belongs or wants to belong to a bigger group. In the pursuit of identity, however, people often get trapped and labelled solely by the identity of the group they want to belong to, ending in losing their personal identity and being nothing more than robotic voices.
“Ode to Identity.” Deep learning art, 2017.
Content basis for “Ode to Identity.”
Deep learning on a minimal sketch.
“Passion. AI Muse” Deep learning art, 2017.
Neural style transfer has been widely used to transfer artist painting style to photographs creating new pieces of art. The opposite direction has motivated this project, and the latter investigates the ability to create photorealistic images. Can simple sketches be brought to life? Using only a minimalistic sketch, a photograph of a person and a neural style transfer algorithm, the machine can bring simple illustrations into life transferring the human characteristics, for example, the eyes, ears, mouth. The neural style transfer algorithm used in this work is trained on the pre-trained VGG-19 convolutional neural network and based on Leon A. Gatys' paper, "A Neural Algorithm of Artistic Style". "Orestes. Erinyes. Greek mythology." is the outcome of this algorithm given a digital drawing and a person's photo, which are both artworks of mine.
In Greek mythology, Orestes was the son of Agamemnon, king of Mycenae, who was murdered by his mother's lover when he returned from Troy. Orestes killed his mother Clytemnestra and her lover, causing the wrath of Erinyes, the goddesses of vengeance and retribution. A torturing madness was inflicted upon him for the crime of matricide by exposing him to his internal feelings of guilt and regret, showing how powerful these emotions can be; nothing more tormenting than guilt.
Orestes. Erinyes. Greek Mythology. Deep learning art, 2019.
Content basis for “Orestes. Erinyes. Greek Mythology.”
