The Happiness is a subjective human experience, and the term "Happiness" is at the same time a word of complex interpretation. However, studies show strong correlation between the population from a specific geographical area and some socio-economic parameters. According to the World Happiness Report (WHR), specific indicators are employed today for an objective interpretation of the personal, economic and social well-being of the resident population in a given country. The indicator called "Life Ladder (LL)", the numerical parameter that refers to well-being, is included among the indicators of a country for progress. This thesis proposes for the first time the development of an artistic installation based on the use of innovative Deep Learning techniques trained of data by World Happiness reports to generate visual experiences that express the happiness ladder identified by a set of happiness indicators set by the attenders. We use Regression as a method to set up a trained environment on the eleven indicators of happiness proposed by the WHR. The system can generate the LL from other indicators entered by the user who is interacting, in line with the well-being levels experimentally detected throughout the globe. We use the inferred LL as input to a Instance Conditioned Generative Adversarial Network (IC-GAN) to generate images and we conduct the human-machine interaction through the use of custom user images as an instance conditioning the image generation. Finally, we use an Enhanced Super Resolution Generative Adversarial Network (ESRGAN) as a technique to provide an increase of the resolution and detail to the generated image as well as to add texture and depth referable to an image drawn by an expert hand. Through a perceptual test we evaluate the positive correlation found between the calculated individual data and their representation through the image subsequently generated.
La felicità è una esperienza umana soggettiva e il termine “felicità” è allo stesso tempo una parola di complessa interpretazione. Tuttavia, studi dimostrano una forte correlazione tra la popolazione proveniente da una specifica area geografica e alcuni parametri di tipo socioeconomico. Secondo il World Happiness Report (WHR), specifici indicatori vengono utilizzati oggi per una interpretazione oggettiva del benessere personale, economico, sociale della popolazione residente in uno specifico paese. L’indicatore denominato “Life Ladder (LL)”, il parametro numerico che si riferisce al benessere, è inserito tra gli indicatori che definiscono il progresso generale di un paese. Questa tesi propone per la prima volta lo sviluppo di una installazione artistica basata sull’utilizzo di tecniche innovative di Deep Learning allenate sui dati del World Happiness report per la generazione di esperienza visive che esprimano il livello di felicità identificato da un insieme di indicatori definiti dall’utente. Utilizziamo la Regressione come metodo per configurare un ambiente allenato sugli undici indicatori di felicità proposti dal WHR. Il sistema è abile a generare il LL da altri indicatori inseriti dall’utente che interagisce col sistema, in linea con i livelli di benessere sperimentalmente rilevati in tutto il mondo. Utilizziamo il LL predetto come input per una Instance Conditioned Generative Adversarial Network (IC-GAN) per generare immagini e conduciamo l’interazione essere umano – macchina attraverso l’utilizzo di immagini personalizzate dall’utente come istanza condizionante la generazione. Infine, utilizziamo una Enhanced Super Resolution Generative Adversarial Network (ESRGAN) come tecnica per fornire un incremento della risoluzione e dettaglio all’immagine generata e aggiungere texture e profondità riferibili a un’immagine disegnata da una mano esperta. Attraverso un test percettivo analizziamo la correlazione positiva trovata tra il dato individuale calcolato e la sua rappresentazione tramite l’immagine successivamente generata.
Development of an artistic installation on world happiness exploiting deep learning techniques for conditioned visual experience generation
Cantoni, Corrado
2021/2022
Abstract
The Happiness is a subjective human experience, and the term "Happiness" is at the same time a word of complex interpretation. However, studies show strong correlation between the population from a specific geographical area and some socio-economic parameters. According to the World Happiness Report (WHR), specific indicators are employed today for an objective interpretation of the personal, economic and social well-being of the resident population in a given country. The indicator called "Life Ladder (LL)", the numerical parameter that refers to well-being, is included among the indicators of a country for progress. This thesis proposes for the first time the development of an artistic installation based on the use of innovative Deep Learning techniques trained of data by World Happiness reports to generate visual experiences that express the happiness ladder identified by a set of happiness indicators set by the attenders. We use Regression as a method to set up a trained environment on the eleven indicators of happiness proposed by the WHR. The system can generate the LL from other indicators entered by the user who is interacting, in line with the well-being levels experimentally detected throughout the globe. We use the inferred LL as input to a Instance Conditioned Generative Adversarial Network (IC-GAN) to generate images and we conduct the human-machine interaction through the use of custom user images as an instance conditioning the image generation. Finally, we use an Enhanced Super Resolution Generative Adversarial Network (ESRGAN) as a technique to provide an increase of the resolution and detail to the generated image as well as to add texture and depth referable to an image drawn by an expert hand. Through a perceptual test we evaluate the positive correlation found between the calculated individual data and their representation through the image subsequently generated.File | Dimensione | Formato | |
---|---|---|---|
M.Sc. Thesis - Corrado Cantoni.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
10.82 MB
Formato
Adobe PDF
|
10.82 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/187919