What makes an image photorealistic and how to pinpoint and understand how our mind interprets different elements in an image conditions? It is proposed that the phrase "imperfect makes perfect" is the key for the photorealistic goal in today’s 3D. There is a review of all the elements for the creation of one perfect image, such as Global Illumination, Anti-Aliasing and also a basic review of photography, how a scene is set up, color temperature and the nature of the real light. To put different theories to a test, the common three dimensional software 3D Studio Max was used with the V-Ray renderer. On a field trip to IKEA communications, we were assigned a project of a room scene containing a kitchen, with a finished scene model. A kitchen was created and experimented to reach a result where there is no visible difference between a computer generated image and the photography. Our result was not what we had hoped for due to many problems with our scene. We ourselves see this as a first step toward a scientific explanation to photorealism and what makes something photorealistic.

[pt] A principal meta da computação gráfica é a síntese de imagens. Essas imagens podem ser tanto geradas por computador quanto ser resultado de manipulação digital de fotografias. Diferentes métodos para captura de imagens e fotografia digital mudaram a importância da imagem digital. Em relação a imagens geradas por computador, a busca por imagens mais realistas é importante para a indústria de filmes, de desenvolvimento de jogos entre outras. Uma das maiores revoluções na computação gráfica atual está relacionada a imagens High Dynamic Range. Essas imagens representam o próximo nível em termos de representação de imagens, uma vez que seus valores são verdadeiramente proporcionais às condições de iluminação em uma cena e são capazes de codificar a dynamic range encontrado no mundo real, fato impossível para as imagens tradicionais que apresentam 24 bits por pixel. Quando imagens high dynamic range são utilizadas para codificar as condições de iluminação em uma cena, elas são chamadas mapas de radiância ou mapas de iluminação. O foco principal dessa dissertação é mostrar técnicas de renderização em tempo real utilizando mapas de iluminação. Técnicas conhecidas como Image Based Lighting. Esse trabalho apresenta os conceitos de imagens high dynamic range, seus fundamentos físicos na teoria da luz, uma série de trabalhos importantes em manipulação dessas imagens e uma discussão sobre o pipeline de aplicações em tempo real que utilizam high dynamic range. Finalmente são apresentadas as técnicas para utilização de mapas de iluminação em alta precisão em tempo real. / [en] In 1997, the seminal work by Paul Debevec and Jitendra Malik on the generation of HDR (High Dynamic Range) images, from ordinary LDR (Low Dynamic Range) cameras, facilitated the generation of light probes enormously. In consequence, this caused a boom of works on the rendering of objects with images of light from the real world, which is known as Image- Based Lighting. The present dissertation aims to study this new area, trying to situate itself in the question of real-time compositing of synthetic objects in real images. This dissertation proposes a real-time rendering pipeline for 3D games, in the simple case of static scenes, adapting the non-real-time technique presented by Paul Debevec in 1998. There is no written work about this adaptation in the literature, although exists some reference to developments done by graphics card manufacturers in this direction. Also this dissertation presents an experiment with diffuse objects. Moreover, the author gives ideas towards the solution of shadow problems for diffuse objects.

L'un des buts principaux de la synthèse d'image est de générer une image en adéquation avec les attentes de l’utilisateur. Pour ce faire, l'utilisateur doit passer par plusieurs étapes. La première, dite « Moteur de rendu », a pour but de simuler de façon précise les différentes interactions lumineuses avec les objets d'une scène 3D. A l'issue de cette étape, l'utilisateur aura à sa disposition une image décrite par des grandeurs photométriques. Ensuite, l'utilisateur procède à une seconde étape, dite « Post-production », où l'utilisateur effectue différents traitements sur l’image générée. Afin que l’image finale soit en adéquation avec les attentes de l’utilisateur, ce dernier peut modifier la scène 3D ou les différents paramètres utilisés par chaque étape. Les travaux de recherche menés dans le cadre de la thèse se sont focalisés sur deux axes principaux : l’accélération de la génération d'images de synthèse et le développement d'outils, proposés à l'utilisateur, lui permettant de satisfaire ses attentes. Concernant le premier axe, nous avons travaillé sur les techniques de synthèse d'image permettant de générer des images de synthèse de haute qualité. Ces techniques s’appuient, le plus souvent, sur un processus stochastique qui construit de façon aléatoire des chemins de lumière. Cependant, dans certains cas de figure, il est difficile de construire de façon efficace ces chemins. C'est le cas des milieux participants (fumée, feu, etc.) pour lesquels un grand nombre d'interactions lumière/matière doit être pris en compte, ce qui est très coûteux en temps de calcul. Pour réduire ces temps de calcul, nous proposons une adaptation d'une approche de rendu discrète exploitant la puissance de calcul des cartes graphiques. Cependant, comme ce genre de techniques présentent de nombreuses limitations, nous avons développé une seconde technique basée sur le trace de photon. Par ailleurs, dans des scènes complexes, il est difficile de trouver des chemins contributifs. Pour cette raison, nous avons utilisé une approche, basée sur Metropolis-Hasting, qui permet d'explorer localement l'espace des chemins contributifs. En revanche, avec ce type d'approche, il est difficile de contrôler la répartition de l'erreur sur le plan image. C'est pourquoi, nous proposons une nouvelle approche permettant de mieux répartir l'erreur relative sur le plan image. Dans le second axe de travail, nous nous sommes intéressés à l’étape de « Post-production ». En effet, nous avons développé une nouvelle technique d'estimation de l'illuminant de référence. Connaître cet illuminant est important lors d’opérations manipulant l'espace couleur utilisé. Enfin, nous proposons une technique capable de déterminer automatiquement la configuration des sources de lumière dans le but de générer une image répondant aux attentes de l'utilisateur. / One of the main goal in computer graphics is to generate an image that matches the user intent. To do that, the user has to go through several steps. The first step, named « Rendering engine », aims to precisely simulate light interactions with the objects of a 3D scene. At the end of this step, an image is generated. This image is represented by photometric values. Then, the user moves on to the second step, named « Post-production », where she/he applies several transformations to the computer generated images. To make the final image be in line with her/his expectations, the user can modify the 3D scene or change parameter values used throughout the different steps. Two main research avenues are investigated: acceleration of the generation of computer generated images and the development of user assistance tools allowing to satisfy the user's intent. First, we have developed computer graphics algorithms that generate high quality images. These techniques often rely on a stochastic process. They randomly construct light paths. However, in some particular setup, it turns out that these algorithms are inefficient. This is the case when rendering participating media for which a huge amount of light interactions is needed. These interactions entail a costly computing time. In order to reduce the rendering time, we have proposed a new discrete approach that runs on the GPU. However, there exist several limitations with this type of technique. To overcome these limitations, we have developed a second approach based on progressive photon mapping. Furthermore, in complex scenes, it is difficult to find valid light paths. This is why, our algorithm is based on Metropolis-Hasting. This type of technique allows to explore locally the path space but still have several drawbacks. The main drawback is that the algorithm does not distribute evenly the error over the image space. We have proposed a novel approach to address this issue. In the second part of this PhD, we are interested in the "post-production" step. A new technique has been developed to estimate the main illuminant in a scene. Knowing this main illuminant is crucial for color transformations. Moreover, we have developed a technique that optimizes automatically the lighting setup in a 3D scene to meet the user's expectations.

Nous proposons une méthode pour placer automatiquement des sondes dans une scène par minimisation d’une fonction d’erreur. Nous guidons les sondes vers les sites d’échantillonnage optimaux en appliquant la descente de gradient à une fonction d’erreur qui représente la similarité entre la structure en construction et un ensemble de référence. En utilisant la pondération inverse à la distance comme fonction interpolante, nous avons construit avec fiabilité des ensembles de sondes dans trois scènes. En comparant nos résultats avec ceux produits par un ensemble de sondes de référence placées sur une grille régulière, nous atteignons théoriquement notre objectif dans une des trois scènes, où nous obtenons des valeurs d’erreur inférieures à la référence avec beaucoup moins de sondes. Nous avons eu des succès partiels dans les autres scènes, selon le nombre d’échantillons utilisés. / Diffuse global illumination within a 3D scene can be approximated in real time using irradiance probes. Probe placement typically relies on significant human input, and final quality of the approximation is often left to the subjectivity of a lighting artist. As demand for realism in rendering increases, the need to enhance the quality of such approximations is greater. We propose a method to automatically place probes in a scene by minimizing an error function. We guide probes to optimal sampling locations by applying gradient descent to an error function that represents similarity between our interpolated results and reference irradiance values. Using weighted nearest neighbour interpolation, we were able to reliably construct probe sets with minimal input in three scenes. Comparing our results to those produced by a set of probes placed on a 3D grid, we were theoretically successful in one scene, in which we could obtain lower error values with fewer probes. We obtained partial success in the other scenes, depending on the number of samples used.

Les images de synthèse sont présentes à travers un grand nombre d'applications tel que les jeux vidéo, le cinéma, l'architecture, la publicité, l'art, la réalité virtuelle, la visualisation scientifique, l'ingénierie en éclairage, etc. En conséquence, la demande en photoréalisme et techniques de rendu rapide ne cesse d'augmenter. Le rendu réaliste d'une scène virtuelle nécessite l'estimation de son illumination globale grâce à une simulation du transport de lumière, un processus coûteux en temps de calcul dont la vitesse de convergence diminue généralement lorsque la complexité de la scène augmente. En particulier, une forte illumination indirecte combinée à de nombreuses occlusions constitue une caractéristique globale de la scène que les techniques existantes ont du mal à gérer. Cette thèse s'intéresse à ce problème à travers l'application de techniques d'analyse de formes pour le rendu 3D.Notre principal outil est un squelette curviligne du vide de la scène, représenté par un graphe contenant des informations sur la topologie et la géométrie de la scène. Ce squelette nous permet de proposer de nouvelles méthodes pour améliorer des techniques de rendu temps réel et non temps réel. Concernant le rendu temps réel, nous utilisons les informations géométriques du squelette afin d'approximer le rendu des ombres projetés par un grand nombre de points virtuels de lumière représentant l'illumination indirecte de la scène 3D.Pour ce qui est du rendu non temps réel, nos travaux se concentrent sur des algorithmes basés sur l'échantillonnage de chemins, constituant actuellement le principal paradigme en rendu physiquement plausible. Notre squelette mène au développement de nouvelles stratégies d'échantillonnage de chemins, guidés par des caractéristiques topologiques et géométriques. Nous adressons également ce problème à l'aide d'un second outil d'analyse de formes: la fonction d'ouverture du vide de la scène, décrivant l'épaisseur locale du vide en chacun de ses points. Nos contributions offrent une amélioration des méthodes existantes and indiquent clairement que l'analyse de formes offre de nombreuses opportunités pour le développement de nouvelles techniques de rendu 3D / Nowadays, computer generated images can be found everywhere, through a wide range of applications such as video games, cinema, architecture, publicity, artistic design, virtual reality, scientific visualization, lighting engineering, etc. Consequently, the need for visual realism and fast rendering is increasingly growing. Realistic rendering involves the estimation of global illumination through light transport simulation, a time consuming process for which the convergence rate generally decreases as the complexity of the input virtual 3D scene increases. In particular, occlusions and strong indirect illumination are global features of the scene that are difficult to handle efficiently with existing techniques. This thesis addresses this problem through the application of discrete shape analysis to rendering. Our main tool is a curvilinear skeleton of the empty space of the scene, a sparse graph containing important geometric and topological information about the structure of the scene. By taking advantage of this skeleton, we propose new methods to improve both real-time and off-line rendering methods. Concerning real-time rendering, we exploit geometric information carried by the skeleton for the approximation of shadows casted by a large set of virtual point lights representing the indirect illumination of the 3D scene. Regarding off-line rendering, our works focus on algorithms based on path sampling, that constitute the main paradigm of state-of-the-art methods addressing physically based rendering. Our skeleton leads to new efficient path sampling strategies guided by topological and geometric features. Addressing the same problem, we also propose a sampling strategy based on a second tool from discrete shape analysis: the opening function of the empty space of the scene, describing the local thickness of that space at each point. Our contributions demonstrate improvements over existing approaches and clearly indicate that discrete shape analysis offers many opportunities for the development of new rendering techniques

This thesis is dedicated to ray tracing based rendering methods, primarily the original ray tracing. Besides introducing a brief historical overview of algorithms from the family, it presents all the essential tools, techniques and physics needed for designing a rendering application in detail. A significant part of the document consists of an implementation of a photorealistic rendering application for interactive graphics 3D virtual scenes. The focus is on rendering without using any additional model information. The thesis includes descriptions and explanations of specific problems and their solutions.

The aim of this work is to create a simple raytracer with IPP library, which will use the frameless rendering technique. The first part of this work focuses on the raytracing method. The next part analyzes the frameless rendering technique and its adaptive version with focus on adaptive sampling. Third part describes the IPP library and implementation of a simple raytracer using this library. The last part evaluates the speed and rendering quality of the implemented system.

Photorealistic rendering has a wide variety of applications, and so there are many rendering algorithms and their variations tailored for specific use cases. Even though practically all of them do physically-based simulations of light transport, their results on the same scene are often different - sometimes because of the nature of a given algorithm or in a worse case because of bugs in their implementation. It is difficult to compare these algorithms, especially across different rendering frameworks, because there is not any standardized testing software or dataset available. Therefore, the only way to get an unbiased comparison of algorithms is to create and use your dataset or reimplement the algorithms in one rendering framework of choice, but both solutions can be difficult and time-consuming. We address these problems with our test suite based on a rigorously defined methodology of evaluation of light transport algorithms. We present a scripting framework for automated testing and fast comparison of rendering results and provide a documented set of non-volumetric test scenes for most popular research-oriented render- ing frameworks. Our test suite is easily extensible to support additional renderers and scenes. 1

This thesis deals with practical implementation of photon mapping algorithm. To achieve better results some basic and some more advanced methods of global illumination has been examined. These time demanding algorithms are often practically unusable and their further optimization is necessary. Optimized ray tracer is essential for practical implementation. Computing diffuse interreflection by Monte Carlo sampling is also very time demanding operation. Therefore it is appropriate to use it along with proper interpolation.

La technique de tracé de chemins est la méthode Monte Carlo la plus populaire en infographie pour résoudre le problème de l'illumination globale. Une image produite par tracé de chemins est beaucoup plus photoréaliste que les méthodes standard tel que le rendu par rasterisation et même le lancer de rayons. Mais le tracé de chemins est coûteux et converge lentement, produisant une image bruitée lorsqu'elle n'est pas convergée. De nombreuses méthodes visant à accélérer le tracé de chemins ont été développées, mais chacune présente ses propres défauts et contraintes. Dans les dernières avancées en apprentissage profond, en particulier dans le domaine des modèles génératifs conditionnels, il a été démontré que ces modèles sont capables de bien apprendre, modéliser et tirer des échantillons à partir de distributions complexes. Comme le tracé de chemins dépend également d'un tel processus sur une distribution complexe, nous examinons les similarités entre ces deux problèmes et modélisons le processus de tracé de chemins comme un processus génératif. Ce processus peut ensuite être utilisé pour construire un estimateur efficace avec un réseau neuronal afin d'accélérer le temps de rendu sans trop d'hypothèses sur la scène. Nous montrons que notre estimateur neuronal (NPPP), utilisé avec le tracé de chemins, peut améliorer les temps de rendu d'une manière considérable sans beaucoup compromettre sur la qualité du rendu. Nous montrons également que l'estimateur est très flexible et permet à un utilisateur de contrôler et de prioriser la qualité ou le temps de rendu, sans autre modification ou entraînement du réseau neuronal. / Path tracing is one of the most popular Monte Carlo methods used in computer graphics to solve the problem of global illumination. A path traced image is much more photorealistic compared to standard rendering methods such as rasterization and even ray tracing. Unfortunately, path tracing is expensive to compute and slow to converge, resulting in noisy images when unconverged. Many methods aimed to accelerate path tracing have been developed, but each has its own downsides and limitiations. Recent advances in deep learning, especially with conditional generative models, have shown to be very capable at learning, modeling, and sampling from complex distributions. As path tracing is also dependent on sampling from complex distributions, we investigate the similarities between the two problems and model the path tracing process itself as a conditional generative process. It can then be used to build an efficient neural estimator that allows us to accelerate rendering time with as few assumptions as possible. We show that our neural estimator (NPPP) used along with path tracing can improve rendering time by a considerable amount without compromising much in rendering quality. The estimator is also shown to be very flexible and allows a user to control and prioritize quality or rendering time, without any further training or modifications to the neural network.