Evaluating user preference when applying mipmap LOD in shadow covered textures

(1)

Bachelor of Science in Digital Game Development March 2020

Evaluating user preference when

applying mipmap LOD in shadow

covered textures.

Jonas Berggren

(2)

This thesis is submitted to the Faculty of Computing at Blekinge Institute of Technology in partial fulﬁlment of the requirements for the degree of Bachelor of Science in Digital Game Development. The thesis is equivalent to 10 weeks of full time studies.

The authors declare that they are the sole authors of this thesis and that they have not used any sources other than those listed in the bibliography and identiﬁed as references. They further declare that they have not submitted this thesis at any other institution to obtain a degree.

Contact Information: Author(s):

Jonas Berggren

E-mail: Jobn16@student.bth.se

University advisor:

Diego Navarro, Petar Jercic Department of Computer Science

Faculty of Computing Internet : www.bth.se

Blekinge Institute of Technology Phone : +46 455 38 50 00

(3)

Abstract

Background. Shadow mapping is a method that is used for generating and imitat-ing shadows in 3D-spaces. This technique has been used in the entertainment media industry in the form of games, movies and 3D renderings of environments to create a more realistic experience for consumers. Shadow mapping is not a perfect technique, and is performance costing on the GPU; however, some methods save performance by reducing the complexity of geometrical shapes and textures depending on the distance between observer and object. These techniques are based on that the ob-server will not notice the complexity reductions; can the usage of such methods be extended to textures covered in shadows without any consequences in the aspect of visual appearance and preference?

Objectives. This thesis aims to examine if there is a possibility to extend the usage of LOD techniques to shadowed textures and to analyze individuals’ preferences of texture variants that are covered in shadows. Additionally, proposing the method of lowering texture resolution by using DirectX:s sampler data type, which is conﬁg-urable to increase the level of details with mipmapping when sampling textures. Methods. This document presents a user study using the Two-alternative forced

choice method and PsychoPy application to create a visual test. The visual test was

conducted in a controlled and observed environment with volunteering participants. The objective of the visual test was to go through several sets of different images, and to choose which image of each set that was preferred. The stimulus was repeated with the initial images fading in and out slowly to prevent carry over effects. After the test participants were asked to fill out a questionnaire. The questionnaire assessed if they noticed any differences within the shadows, and if they had any additional thoughts about the experiment.

Results. The results from the study were then evaluated through a binomial test that yielded that there was no statistically signiﬁcant diﬀerence in preference between the lowered texture resolution in shadows and normal texture resolution in shadows. Separately evaluating the environments showed that there was a preference for shad-owed low-resolution textures in environments that were dark. The environments with high illumination had more varied results. There were 17 participants that volun-teered in the test and were ranging from the ages 18 to 29.

Conclusions. With the results presented it was shown that the shadowed low-resolution textures were preferred in environments with low illumination. This sug-gests that the proposed concept method is better suited for similar environments. However, several factors may have aﬀected the results. Factors such as images being too dark, the lack of exaggerated images, images fading in and out too fast, few participants, more partaking will assure there is less chance for bias.

(4)

(5)

Acknowledgments

The author would like to thank Diego Navarro for his help, feedback, and guidance in developing the thesis. The author would also like to thank Petar Jerčić for his supervision and all participants for helping with gathering study data.

(6)

(7)

List of Figures

1.1 Example of shadow mapping in a 3D graphics renderer. . . 1 2.1 Example: Image sizes from mipmapping. From left to right: 1024x1024,

512x512, 256x256, 128x128, 64x64. . . 4 3.1 Screenshot of a scene that was used in the experiment in Autodesk

Maya. . . 8 3.2 A zoomed in image of the castle scene. Left image is a scene rendering

with the proposed concept sampler, and the right image is a rendering of the scene with a normal sampler. . . 9 3.3 A choice in the visual experiment. The scene is a dark castle

envi-ronment. Left image is Normal-resolution textures in shadow. Right image is Low-resolution textures in shadow sampled by the proposed concept sampler. . . 10 3.4 A choice in the visual experiment. The scene is a bright environment

that has some depth. Right image is Normal-resolution textures in shadow. Left image is Low-resolution textures in shadow sampled by the proposed concept sampler. . . 10 4.1 Graph showing the total preferences of visual appearance of textures

that are covered in shadows. "Standard method" representing Normal-resolution textures in shadows and "Proposed Concept" representing low-resolution textures in shadows using the proposed concept method. 13 4.2 Graph showing the total preferences of visual appearance of textures

that are covered in shadows in each sets of scenes. "Standard method" representing Normal-resolution textures in shadows and "Proposed Concept" representing low-resolution textures in shadows using the proposed concept method. . . 13 4.3 One of the images used in sets 1 - 3; displaying the dark castle

envi-ronment with the proposed sampler concept. . . 14 4.4 Graph showing the total preferences of visual appearance of textures

that are covered in shadows in the dark castle environment. "Stan-dard method" representing Normal-resolution textures in shadows and "Proposed Concept" representing low-resolution textures in shadows using the proposed concept method. . . 14

(10)

4.5 Graph showing the preferences of visual appearance of textures that are covered in shadows in the 3 sets that contained images of the dark castle environment. "Standard method" representing Normal-resolution textures in shadows and "Proposed Concept" representing low-resolution textures in shadows using the proposed concept method. 15 4.6 One of the images used in sets 4 - 7; displaying the small bright

envi-ronment with the proposed sampler concept. . . 15 4.7 Graph showing the total preferences of visual appearance of textures

that are covered in shadows. "Standard method" representing Normal-resolution textures in shadows and "Proposed Concept" representing low-resolution textures in shadows using the proposed concept method. 16 4.9 One of the images used in sets 8 - 11; displaying the large bright

environment with the proposed sampler concept. . . 17 4.10 Graph showing the total preferences of visual appearance of textures

that are covered in shadows. "Standard method" representing Normal-resolution textures in shadows and "Proposed Concept" representing low-resolution textures in shadows using the proposed concept method. 18 A.1 Consent form presented to all participants. . . 29 A.2 Experiment description presented alongside the consent form. . . 30

(11)

Chapter 1 Introduction

Entertainment industries, specifically the gaming industry in present day often aims for creating an authentic and immersive environment for the players to experience and explore. There are many ways to create a more immersive environment by simulating realism in games. One of the methods that is frequently used is shadow mapping, which is a technique that simulates shadows in 3D environments in real-time [15]. Figure 1.1 demonstrates this technique. This is a method that grants more depth and better visuals to a 3D surrounding at the cost of performance. This method is not without complications either, some of them include aliasing artefacts, filter-ing problems, peter pannfilter-ing artefacts and shadow acne. But there are techniques made to counteract such problems [1]. Shadow mapping operates by rendering the scene from the perspective of the light and storing the depth values of each fragment point in the scene into a 2D texture. This value will be compared with the rendered depth values of fragment points from the perspective of the viewer, which has been converted to the coordinate space of the light. If the depth value is not the same from both perspectives, then the point is occluded by an object and is covered in shadow. Shadow mapping is a method that computes shadows in real-time, and is a costly process because many samples are needed to be taken from an area that the light-source illuminates for the term of the rendering equation [9].

Figure 1.1: Example of shadow mapping in a 3D graphics renderer.

(12)

2 Chapter 1. Introduction

There are techniques that can be implemented to reduce the costly consequence of such methods, one of the procedures in computer graphics is called Level of De-tail (LOD). The idea of LOD techniques is to reduce the amount of workload in the graphics pipeline stages by decreasing the complexity of geometrical objects and texture details depending on the distance between them and the observer.

However, seeing as the graphics quality aﬀects the playability and immersion of video games through realism, sensory appeal, and satisfaction [13]. From the viewpoint of a video game player, it would be preferred to have good texture quality to improve immersion and enjoyment when playing. If LOD techniques improves performance by lowering the resolution of textures at a greater distance and that the details are not as perceivable as it is close up. Can performance be improved further by com-bining an extended use of LOD techniques and shadow mapping to lower texture resolution in shadows more than those not covered by shadow? This seems possible considering that the human eye discerns fewer details in the dark. This is because the eye converts reﬂected light into a nerve signal that is sent to the brain to produce optical imagery of what it sees. [11].

One can possibly beneﬁt and use the information to reduce the visual quality appear-ance of textures that are covered in shadows, or in a dark environment with extended usage of LOD techniques without any larger repercussions. By those means, there presumably is a possibility to extend the usage of other performance saving methods that aﬀect the visual appearance. However, this is not certain, thus needs to be studied.

This thesis intends to analyze if individuals have any preference for visual appearance between shadowed textures with low-resolution and texture with normal-resolution, using a LOD technique to lower the resolution of only shadowed textures. This is to propose whether a concept such as this can be used by others to possibly reduce the costing of shadow mapping through other means than conﬁguring the shadow mapping technique. This will be done by having participants study pictures of the same scene, with diﬀerences in texture resolution in shadows, and then choose which one they prefer the most in the aspect of visual appearance.

1.1 Problem and Aim

The LOD methods are built to decrease the performance cost on rendering objects or images that we cannot perceive clearly, and with how the human eye revolves around how much light that gets absorbed to develop an image [11]. How can one extend the usage for LOD techniques that reduce the size of the images of textures covered in shadows produced by shadow mapping without any observer noticing larger dis-parities in texture quality and aﬀecting their preference?

(13)

1.2. Research Questions 3

1.2 Research Questions

This study will focus on the following research questions:

• Which variant do video game players prefer between shadowed low resolution

textures and shadowed normal resolution textures?

• In what situations is it better to use shadowed low resolution textures?

1.3 Hypothesis and Expected Outcome

One of two hypotheses for this thesis will be chosen and reject the other through analyzing experiment data in this thesis. The hypotheses are:

Alternative hypothesis. There is a noticeable preference between shadowed low

resolution textures and both normal-resolution textures that are either covered in shadow or lit up by light.

Null hypothesis. There is no noticeable preference between shadowed low

reso-lution textures and both normal-resoreso-lution textures that are either covered in shadow or lit up by light.

The anticipated outcome for the user study tests will be that the users might be able to notice a small discrepancy between both images of the shadowed-textures, in addition, it is expected that the participants will prefer the low-resolution tex-ture version. It is anticipated that the low-resolution variant will be useful in any situation.

(14)

Chapter 2 Related Work

Shadow mapping generation in computer graphics is a spectrum in which there has been a lot of improvement and research during the years. It was first introduced by Williams[15]. In the article he proposed a method to gauge shadows in 3D scenes. As previously mentioned this method is not without its visual problems. Several ar-ticles have proposed experiments and explored different usage of methods of filtering in real-time and anti-aliasing [9, 6] methods to optimize shadow mapping.

Mipmapping is a LOD technique that was introduced by Williams [16]. Formally

called “Pyramidal parametrics”, uses the process of pre-generating a string of tex-tures from one texture. Each texture generated is smaller in height and width by 2(n)_{of the parent image in the string. If the original texture is 1024x1024 pixels, the}

next texture in the string will be 512x512, the subsequent texture will be 256x256. This process goes until the images size is at the maximum LOD level, as can be seen in Figure 2.1. Mipmapping have the capability to interpolate textures between two different levels and utilize pre-filtering whenever a textures is viewed from a distance. This method is used to lower details in images that are far away from cameras in 3D scenes to improve rendering speed, reduce aliasing artifacts, and save cache performance and data transference [16]. This is not without a negative side since mipmapping performs all of this at the expense of using more memory per textures that is loaded in [4]. This technique also lowers the amount of texels [3] for each level it rises, which means that there is less information to pass through during real-time that will not be used for much. But foremost it is implemented for its ability to maintain texture definitions of textures when further away from camera, thus increasing visual qualities of textures that are far away.

Figure 2.1: Example: Image sizes from mipmapping. From left to right: 1024x1024, 512x512, 256x256, 128x128, 64x64.

(15)

5

A comparative study made by Maisch et.al., studied and compared the differ-ence between various mipmap modification schemes developed to support feature-preserving rebuilding during rendering [10]. The study discusses the shortcomings of mipmapping and introduced techniques that addressed the problems. The study show performance measurements of the different methods applied and not applied to the mipmap rendering. The authors wraps the study up by stating the conclusion that their approaches can be applied as an improvement to mipmapping. However, they also discusses about the strengths and flaws of each approach.

Investigating possibilities of combining mipmapping and shadow mapping led to a paper written by Donnelly et.al., [8] in which they proposed a new real-time shad-owing algorithm that instead of only storing depth data also stores the mean value and mean-squared of a distribution of depths. Their implementation of shadow map-ping is called "variance shadow mapmap-ping". This variant of shadows mapmap-ping utilized hardware mipmapping, trilinear and 16x anisotropic ﬁltering to create better ﬁltered shadows [8].

Mipmapping has also been used for creating a GPU-based height map intersection technique that is based on hierarchical data structure. This technique was suggested by Tevs et.al., and has shown to be fast and accurate in comparison to other algo-rithms based on pre-computed data [14].

There has been articles that have been combining the techniques of shadow mapping and mipmapping. However, the purpose of those articles have been to improve the aliasing of shadow appearance or improving the rendering performance of shadows [8, 14]. But investigation on articles that combines mipmapping with shadow map-ping to extend the usage of mipmapmap-ping to increase the usage of mipmap LOD levels in shadows has not been found with the keywords and date restrictions used for the retrieval.

(16)

Chapter 3 Methodology

This chapter will focus on introducing the reader on methods, applications, and concepts that were used in this user study. The method that was used for visual ap-pearance preference and stimuli evaluation was two-alternative forced choices, more commonly known as the “2afc method”, which is a method for assessing a subjective preference of an individual through sets of choices. In this method, participants are presented with a question with two alternative options. Only one of the two options contains the participants’ incentive and is forced to make a choice about which one that they prefer.

This chapter will introduce the software used over section 3.1 and hardware used over section 3.2. The explanation of creating 3D scenes over section 3.3, the pro-posed concept sampler method over section 3.4 and the procedure over section 3.5. Lastly, information about participants, ethics and informed consent will be brought up over section 3.6.

3.1 Tools

3.1.1 PsychoPy

PsychoPy is an open-source application that can be used to run, create tests and

experiments in the category of neurosience, psychology and psychophysics. PsychoPy oﬀers the users Build-interface and Coder-interface to build experiments in Python programming language. PsychoPy was created by Peirce et.al. [12].

By using PsychoPy in this thesis it was easier and faster to set up a 2afc experiment. It also allowed to create an experiment that can sort all result immediately after usage. PsychoPy version 3.1.1 was used in this study.

3.1.2 DirectX

Microsoft DirectX is an assemblage of several application programming interfaces (API:s) for conducting tasks that are associated with the multimedia, such as 3D graphics, animation, imagery, sounds and more [2]. DirectX is used in program-ming of games and videos on microsoft platforms. It was ﬁrst released in september 1995 by Microsoft. As mentioned previously it can be used to conﬁgure and control mipmapping with a data structure called sampler description. In this study, the DirectX11 version was used.

(17)

3.2. Apparatus 7

When declaring this structure it can be configured to having a maximum and mini-mum levels of detail for mipmapping. The possibility to set a LOD bias for mipmap-ping is possible too, this means that setting the bias value to a certain number, will make the sampler to sample images in that specific LOD of mipmap. This is the reason why mipmapping and DirectX s sampler will be a benefiting asset to this thesis.

3.1.3 Autodesk - Maya

Autodesk Maya, more commonly known as Maya is a computer application that was

originally developed by Alias Systems Corporation and initially released in february 1998; and is now currently owned and developed by Autodesk. Maya is a 3D computer graphics application that is used to create 3D assets for usage in interactive appli-cations such as video games, animated ﬁlms or TV-series and visual eﬀects. Maya will be used in this study to create assets for a 3D environment, as well exporting scenes for the visual experiment to OBJ format. Autodesk Maya 2018 was used for this study.

3.1.4 Graphical Engine & Assimp

A graphics renderer was developed using C++ as a language and Directx11. Both shadow mapping and mipmapping will be essential parts in this thesis as the entire research circulates these topics, these techniques will be implemented in the graphics renderer. A method to import OBJ models was also developed.

To make an agile development of the OBJ importer, there’ll be an implementation of the Open-Asset-Importer-Lib (Assimp) made by Gessler, et.al., to quickly set up an OBJ importer. Assimp version 4.1.0 was used for importing OBJ throughout this study. With both shadow mapping and mipmapping available an implementation of the approach to solving the problem will be set up to reach the aim goal.

3.2 Apparatus

The testing was performed on a computer that had the following hardware: Graphical Processing Unit: NVIDIA GeForce GTX 1080

Central Processing Unit: Intel(R) Xeon(R) CPU E5-1620 v4 @ 3.50GHz, 3501Mhz, 4 Cores, 8 Logical Processors

Monitor: DELL P2217H Screen Resolution: 1920 x 1080 Refresh rate (Hz): 60Hz

Bit depth: 8-bit Color format: RGB

(18)

8 Chapter 3. Methodology

3.3 Implementing 3D Scenes

The scenes that were used in the visual experiment consist of a majority of 3D models and textures that were made by the author from an earlier occasion in Autodesk Maya. However to have more variation and cases in the visual test more 3D components were to be made. In Maya scenes were constructed up by placing 3D objects around to make an environment as shown in figure 3.1, camera position for each scene snapshots were taken from the camera position in Maya and inserted to the own made 3D renderer. OBJ files that were imported to the renderer were exported from Maya. For this thesis, three different environments were made into different sets of images to be used in the visual experiment. The three different environments were a dim castle scene, a small room with bright textures, a large room with bright textures.

(19)

3.4. Proposed Concept Sampler Method 9

3.4 Proposed Concept Sampler Method

The method utilizes the usage of DirectX:s data types, more specifically the sampler type. Sampling is the process of gathering data from images and maps. These samplers, specifically D3D11_SAMPLER_DESC can be configured to handle data in specific ways such as, how to filter the image data, controlling mipmapping LOD an image should be in, and anisotropy settings to name a few. With a sampler set to always be at an increased level at mipmapping LOD than the usual sampler. Figure 3.2 demonstrates the visual difference between the two samplers. Exclusively using this sampler when sampling texture for the shadowed points during the second pass of the shadow mapping technique will create a simple method to lower textures in shadows.

Figure 3.2: A zoomed in image of the castle scene. Left image is a scene rendering with the proposed concept sampler, and the right image is a rendering of the scene with a normal sampler.

3.5 Procedure

With the entire graphics renderer complete, and all scenes ready; the next step is to proceed with the test. This test will be a visual perception test, that will be con-ducted in a controlled environment. The test will not be timed, and the participants are going to be able to follow through the experiment at their own pace. The test will have several sections; first will be the introduction of the test. This part will inform the participant about the instructions of the test and what the participant can expect from carrying out the test. After been given instructions the participant will go through several sets of different images that can be seen in figures 3.3 and 3.4. Each set has one image that has shadowed low-resolution textures (SLRTs) and one that is with shadowed textures with normal resolution. The participant then has to pick which one for each set that is most preferred in the aspect of visual appearance. This is done by pressing the left-arrow key for the left image, or the right-arrow key for the right image.

(20)

10 Chapter 3. Methodology

Figure 3.3: A choice in the visual experiment. The scene is a dark castle environment. Left image is Normal-resolution textures in shadow. Right image is Low-resolution textures in shadow sampled by the proposed concept sampler.

Figure 3.4: A choice in the visual experiment. The scene is a bright environment that has some depth. Right image is Normal-resolution textures in shadow. Left image is Low-resolution textures in shadow sampled by the proposed concept sampler.

(21)

3.6. Participants 11 The stimulus will be repeated with the initial images fading in and out slowly in four seconds to prevent carry-over eﬀects, such as a memorization pattern in the stimuli presented. This is to make sure that the participants will not choose a scene out of recognition. The LOD of mipmapping will change between the stimuli presented to participants, for each scene, there will be a diﬀerent environment.

After having gone through all the sets the participant will be asked to fill out a questionnaire. These questions serve the purpose of finding if the participants noted any specific difference in textures or other factors that could have affected their choices in the test.

3.6 Participants

There have been studies on human visual cognitive abilities. Specifically in a study made by Boot et.al, [5]. In which there has been the conclusion that video game players (VGPs) visual cognitive abilities are surpassing that of an individual that is not experienced in playing video games (NVGPs) [5]. The areas which it has been tested in before have been visual attention, memory, and executive control [5]. There is also the conclusion that video game players respond quicker than non-video game players to attentional capture tasks [7]. This disclosure affects the study by informing that this research should be enacted on participants that are used to video games. Considering that this kind of individuals is efficient in the areas of visual cognition, meaning they will have it easier to spot visual disparities. Because of this, there is the probability that these individuals can see the difference between texture details of SLRTs and normal-resolution textures that are both lit and shadowed.

The requirement for participants in this test being a VGP, which the inclusion criteria were having previous experience playing 3D video-games of any kind, and frequently partake in activities that involve playing video-games. The deﬁnition of VGP was mentioned in [7] as an individual who played a minimum of three hours per week of action video-games over six months. This requirement was assessed by asking the participants before the test if they frequently engaged in gaming activities and how long they usually played for each activity.

3.6.1 Ethics & Informed Consent

The participants were informed about the purpose of this study and what methods were used to gather data and how much time it might take. Participants also have the possibility to decline participation or leave the study at any time.

Participants were also aware of their conﬁdentiality and were oﬀered the contact information of the person responsible for experimentation. They were briefed on how the data was going to be used and their access to data upon contacting the responsible person. There were no risks involved in partaking the experiment.

(22)

Chapter 4 Results & Data Analyzing

4.1 Results

The data gathered from the visual experimentation came from a total of 17 partici-pants that signed up for participation. The participartici-pants consisted of thirteen males and four females, ranging from the age of 18 to 29. The binomial test was used in this thesis to analyze data from the visual experiment. Results that were gained by performing this data analysis suggests that there was not a statistically signiﬁcant diﬀerence between SLRTs and textures that had a normal resolution in shadow. The entire visual experiment choices yielded that the probability value was 0,099. This implies that the null hypothesis was accepted and the alternative hypothesis was re-jected. This suggests that there is no noticeable preference between the preferences of SLRTs and normal-resolution textures in shadows.

Breaking down the binomial test for all choices, and using binomial testing over three separate scene environments that were used in the visual experiment gives more in-sight into the results. The dark castle environment as seen in figure 3.3 showed that a large portion of the participants preferred SLRTs over normal-resolution textures in shadows. The null hypothesis was accepted with a probability value of 0,096, which suggests that there was no significant statistical difference between both variants. The small light scene had a probability value of 0.353, which shows that the null hypothesis for this environment was accepted. The analysis shows that there is a low preference difference between the SLRTs and normal-resolution textures in this scene. Lastly, the large bright room had a probability value of 0,018, which construes that the null hypothesis was rejected and the alternative hypothesis was accepted. This means that the normal-resolution textures in shadows were preferred over SLRTs in this environment.

4.2 Visual Data

The visual experiment that was conducted had each participant go through a total of 11 different sets, which were composed of two images of the same scene with two different samplers for the textures that were covered by the shadows. The 3 first sets shown in Figure 4.2 consists of scenes of the dark castle environment, sets 4 to 7 consists of scenes of the small bright environment. The sets 8 to 11 consists of scenes from the large bright environment.

(23)

4.2. Visual Data 13

Figure 4.1: Graph showing the total preferences of visual appearance of textures that are covered in shadows. "Standard method" representing Normal-resolution textures in shadows and "Proposed Concept" representing low-resolution textures in shadows using the proposed concept method.

Figure 4.1 shows in total that the most preferred images had been those that had normal-resolution texture in shadows. The pie chart shows that 55% of the choices preferred normal-resolution texture in shadows (Standard Method) and that 45% preferred SLRT images (Proposed Concept Method). Out of the total 176 choices; normal-resolution was chosen 97 times as the most preferred, and low-resolution was chosen 79 times. Figure 4.2 shows the total preference in each scene and how many times each choice was made at each scene of all 11 sets.

Figure 4.2: Graph showing the total preferences of visual appearance of textures that are covered in shadows in each sets of scenes. "Standard method" representing Normal-resolution textures in shadows and "Proposed Concept" representing low-resolution textures in shadows using the proposed concept method.

(24)

14 Chapter 4. Results & Data Analyzing

Analyzing data per environment scene shows that in total the low-resolution texture in the dark castle environment which can be seen in Figure 4.3 was most preferred. Figure 4.4 shows that the low-resolution texture was chosen 29 times and had a percentage of 60% and that the normal-resolution texture in shadows (Standard Method) was chosen 19 times with a percentage of 40% out of total 48 choices. Figure 4.5 shows the preference for visual appearance in each dark castle set. In set 1 the standard method is mostly preferred, while in set 2 it is more even between the choices. Set 3 shows that a large majority preferred the proposed concept.

Figure 4.3: One of the images used in sets 1 - 3; displaying the dark castle environ-ment with the proposed sampler concept.

Figure 4.4: Graph showing the total preferences of visual appearance of textures that are covered in shadows in the dark castle environment. "Standard method" repre-senting Normal-resolution textures in shadows and "Proposed Concept" reprerepre-senting low-resolution textures in shadows using the proposed concept method.

(25)

Figure 4.5: Graph showing the preferences of visual appearance of textures that are covered in shadows in the 3 sets that contained images of the dark castle en-vironment. "Standard method" representing Normal-resolution textures in shadows and "Proposed Concept" representing low-resolution textures in shadows using the proposed concept method.

The small bright room environment which is seen in Figure 4.6, had 64 choices made in total. Figure 4.7 shows that normal-resolution was chosen 34 times as the most preferred and had a percentage of 53%, low-resolution was picked 30 times and had a percentage of 47%. It can be seen in Figure 4.8 that the choices picked in sets 4, 5 and 6 were partially even and in set 7 most preferred the standard method.

Figure 4.6: One of the images used in sets 4 - 7; displaying the small bright environ-ment with the proposed sampler concept.

(26)

In the larger bright room environment, see 4.9, had 64 choices made in total. SLRT was chosen 20 times and had a percentage of 31% and the normal-resolution was picked 44 times, with a percentage of 69%, this can be seen in Figure 4.10. The choices made in sets 8 and 9 are moderately even which can be seen in 4.11 and the sets 10 and 11 in the ﬁgure shows that the participants largely preferred the normal-resolution.

(27)

Figure 4.9: One of the images used in sets 8 - 11; displaying the large bright envi-ronment with the proposed sampler concept.

(28)

4.3 Discarded Data

A fault with the experiment description in not specifying that the participants could take their time watching each set of images during the visual experiment led to that one of the partaking participants rushed through all the sets of images under three minutes. Since the participant did not take time to perceive the images thoroughly, the participant data was annulled and excluded from the results.

(29)

Chapter 5 Analysis and Discussion

5.1 Questionnaire Responses

Multiple responses on the questionnaire showed that the participants noticed tex-ture differences in the shadows, while a small amount did not. Most of the responses expressed confusion and that the only differences in shadows were a darker shade in some images. This type of response was probably caused by the minimal information about what the participants should look for. Few responses mentioned differences in acuity of textures in shadows, which is an effect of the optimizing sampler being in action.

Three participants left responses that indicates that they noticed some diﬀerences between textures, two commented speciﬁcally:

"I perceived the textures to be sharper in some images and more faded in other."

"On one picture there was a very blurry texture where the shadows were." One of the reasons as to why the answers ended up like this could probably be that the lack of explicit instructions given to the participants aﬀected the way they evalu-ated each of the pictures presented to them. This means that there were less quality and content for the visual experiment that could have aﬀected the answers. Another reason could be that the information about that the participant should pick the im-ages that they preferred most from visual appearance was given orally.

5.2 Aﬀecting Variables

Visual data results supports the null hypothesis: There is no preference between SLRTs and both normal-resolution textures that are either covered in shadow or lit up by light. The results showed that the observed diﬀerence between normal sam-pled textures and textures samsam-pled with the proposed concept sampler was reliable enough to say that the viewer preference diﬀerence between both samplers were not large.

(30)

20 Chapter 5. Analysis and Discussion

However; there are several factors that can have aﬀected the gathered data from the visual experiment, such as:

• Images being too dark or fading out too fast. • Diﬀerent darkness levels on the shadows.

• Variety of the stimuli; images that pushed the capabilities of the evaluated method.

• Be more explicit about no time limitations.

• Too few participants, to claim statistical signiﬁcance. • Participants given information orally.

Further investigating with more visual experiments and clearing up these points could give a different result. More testing is needed to receive more detailed and solid data that can be used for future research. A noteworthy point is how differ-ent computer monitors can affect how images are displayed. Performing this test on different computer monitors with differing settings might give different answers, therefore operating the visual experiment on various monitors can give more insight-ful data. Another point that might have affected is an increase size of different scenes. Having more larger scenes with increased amount of differing 3D components and textures to work with.

As mentioned previously; information on that the participants should choose an image that they preferred most from the standpoint of visual appearance. This in-formation was given orally before the test started, which could have aﬀected study since the participant could have misheard or only focused on the instructions paper which only mentions ’preference’. This could be resolved by being more pronounced in the paper that the participants are to choose images by their visual appearance preference.

This visual experiment only resorted to images, how could the answers have diﬀered if this concept sampler method was implemented in a game, movie or a situation that contains a lot of action? Would the participants notice the disparities between SLRTs and normal-resolution textures more clearly, or would it distract the participants to the point which they do not notice the diﬀerence in details? More testing in this area could give more evidence that supports either the null hypothesis or alternative hypothesis more.

Results showed that more than half of the participants preferred the SLRTs in the dim castle environment. In addition, it was shown that there was no notable statis-tical diﬀerence. Especially in comparison to the other two environments. This gives the implication that it is harder for the participants to see a diﬀerence in textures

(31)

5.3. Study Limitations 21 when in a darker surrounding or when the textures that are duller in color. Stimuli sets 3, 8 and 11 which can be seen in Figure 4.2 in chapter 4, contained images with a high LOD level using the proposed concept method.

The textures in those images had a much lower resolution on the textures in the shadows that the other variants. There is a possibility that the images that were in those sets affected the results too. However, results showed that most participants chose the SLRT variant in set 3 which was a set consisting of images of the dark cas-tle scene, further suggests that the participants have a harder time seeing differences in a darker environment. In set 8, which consists of images that showed the small bright scene, had a more varied result in choices. Set 11 which consisted of images showing the large bright scene showed that a large number of participants chose the standard method, indicating that they noticed the difference of texture resolutions. This can be something that one can take advantage of when using the proposed concept method. There is also the probability that the idea of this proposed concept method can be applied by others for additional techniques for reducing performance costs.

Given the results, it is also noted that it is probable that 3D graphics and enter-tainment industries can use this information to further improve the quality of future games. Nevertheless, more testing, data analyzing and observation on this part can be beneﬁcial for drawing a steady conclusion.

5.3 Study Limitations

Due to the nature of this study, a decision was made to create a 3D renderer speciﬁc for this study to have more control over creating the concept sampler method. How-ever, since the study was performed only on one renderer. There is no knowledge or data on how it would appear in other renderers such as Unreal Engine 4 and Unity. To achieve a more generalized verdict of data, studies with this method implemented in other renderers should have been conducted.

5.4 Extending the Usage of Concept Method

Since the results suggest that there is no significant statistical difference between the two variants of texture-resolutions, there might be other fields the proposed concept can apply to. Extending the use of mipmapping to areas such as gaze mapping or eye-tracking could be possible. Since gaze mapping revolves around the idea of accu-mulating information from eye trackers to examine the observed behavior of a user. Maybe one could extend the usage of mipmapping to areas which the user does not usually observe.

It might be plausible to extend the proposed concept to dark areas and dark tex-tures, instead of only shadows to further explore the capabilities of mipmapping. Especially since the dim castle scene showed no signiﬁcant statistical diﬀerence be-tween both variants, and had a larger amount of choices that preferred SLRTs over

(32)

22 Chapter 5. Analysis and Discussion

normal-resolution textures. However, this can involve some risks too. If the method would be extended and then be used in an environment that is dim it might re-duce the overall texture quality of the environment. Which by extension will aﬀect perceived graphical quality.

(33)

Chapter 6 Conclusions and Future Work

6.1 Conclusions

This study presents the inquiry if individuals have any preference between lowered texture details and normal texture details of textures that are covered in shadow or dark, and if there is a possibility one can use this information to apply a method that can save cache performance and rendering speed with the increased use of mipmap-ping LOD on textures covered in shadows.

The research questions brought up followed as:

• Which variant do video game players prefer between shadowed low resolution

textures and shadowed normal resolution textures?

• In what situations is it better to use shadowed low resolution textures?

This study determined the answer to that question by stating that video game players did not prefer either variant significantly more over the other. However, the results showed that the participants preferred the low-resolution variant more in dark en-vironments. The bright environment showed a more varied result of preference; the small bright scene showed no large preference difference between both variants, while the large scene showed a large preference for the normal resolution variant. It was suggested that the proposed concept method was more effective in dark environments and is more suited for such environments. However, it was also shown that it was effective in some cases in a bright environment.

Data produced from a visual test in form of the 2afc method by using PsychoPy application, analyzed by binomial test shows that the observed differences between SLRTs and normal-resolution textures are not convincing enough to say that the no-ticeable difference between both variants is high. However, this study did not include enough participants to convey any statistically significant results. This paper further reflects over complications and points that could have affected the results. Points and complications being that different computer monitors might give affect the vi-sual experiment. Additionally how environments with various lightning can affect what the participants can see, and if a more action-filled 3D environment would give different results. Including the complications that appeared during visual testing such as images being too dark or fading too fast, lacking exaggerated images.

(34)

24 Chapter 6. Conclusions and Future Work

More testing with addressing the complications would bolster and solidify the re-sults. This study progresses by mentioning how this work can be used for future productions in the entertainment media and 3D graphics industries by teaching how humans discern texture details that are covered in shadows or darker environments and how industries can utilize this information to improve quality and performance of future games.

6.2 Future Work

In this study, results have shown that individuals that cooperated did not notice any diﬀerences in texture resolutions of texture that are covered in darkness. This result supports the before mentioned hypothesis and has brought answers to the research questions that were brought forth.

More testing with the complications in mind, larger participant scale, scenes with more diﬀering textures and models would lead to more reliable results. The data, thoughts, and conclusions accumulated from this study can be used by the enter-tainment media or 3D graphics industry to understand more about how humans can discern texture details that are covered in shadow, darker lighting, and settings. Therefore can make the judgment of how they can use the thesis concept to improve the performance and quality of future games by optimization. Additionally with this information examining the possibility of extending the use of mipmapping together with techniques such as gaze mapping to create a method that utilizes the extended usage of mipmapping on parts of the screen that is not usually observed.

(35)

References

[1] Microsoft: Windows Dev Center: common techniques to improve shadow depth maps. https://docs.microsoft.com/en-us/windows/win32/dxtecharts/common-techniques-to-improve-shadow-depth-maps. (Accessed: 2019-05-04).

[2] Microsoft: Windows Dev Center: direct3d 11 graphics. https://docs.microsoft.com/en-us/windows/win32/api/_direct3d11/. (Ac-cessed: 2019-04-12).

[3] Microsoft: Windows Dev Center: introduction to textures in direct3d 11. https://docs.microsoft.com/en-us/windows/win32/direct3d11/overviews-direct3d-11-resources-textures-intro. (Accessed: 2019-04-12).

[4] Microsoft: Windows Dev Center: texture ﬁltering with mipmaps (direct3d 9). https://docs.microsoft.com/en-us/windows/win32/direct3d9/texture-ﬁltering-with-mipmaps. (Accessed: 2019-04-12).

[5] Boot, W. R., Kramer, A. F., Simons, D. J., Fabiani, M., and Gratton, G. The eﬀects of video game playing on attention, memory, and executive control. Acta Psychologica 129, 3 (Nov. 2008), 387–398.

[6] Cerqueira de Farias Macedo, M., Vinicius Teixeira, A., Lopes Apolinario Junior, A., and Apaza Aguero, K. Hard Shadow Anti-Aliasing for Spot Lights in a Game Engine. In 2017 16th Brazilian Symposium

on Computer Games and Digital Entertainment (SBGames) (Curitiba, Nov.

2017), IEEE, pp. 106–115.

[7] Chisholm, J. D., Hickey, C., Theeuwes, J., and Kingstone, A. Re-duced attentional capture in action video game players. Attention, Perception,

& Psychophysics 72, 3 (Apr. 2010), 667–671.

[8] Donnelly, W., and Lauritzen, A. Variance shadow maps. In Proceedings

of the 2006 symposium on Interactive 3D graphics and games - SI3D ’06, ACM

Press, p. 161.

[9] Macedo, M., and Apolinário, A. Improved anti-aliasing for Euclidean distance transform shadow mapping. Computers & Graphics 71 (Apr. 2018), 166–179.

[10] Maisch, S., Skanberg, R., and Ropinski, T. A comparative study of mipmapping techniques for interactive volume visualization. 35–42.

(36)

26 References

[11] Oleari, C., and Simone, G. Standard colorimetry: deﬁnitions, algorithms,

and software. Society of dyers and colourists. John Wiley, Chichester, West

Sussex, 2016. OCLC: 937981097.

[12] Peirce, J., Gray, J. R., Simpson, S., MacAskill, M., Höchenberger, R., Sogo, H., Kastman, E., and Lindeløv, J. K. PsychoPy2: Experiments in behavior made easy. Behavior Research Methods 51, 1 (Feb. 2019), 195–203. [13] Sánchez, J. L. G., Vela, F. L. G., Simarro, F. M., and Padilla-Zea,

N. Playability: analysing user experience in video games. 1033–1054.

[14] Tevs, A., Ihrke, I., and Seidel, H.-P. Maximum mipmaps for fast, accu-rate, and scalable dynamic height ﬁeld rendering. In Proceedings of the 2008

symposium on Interactive 3D graphics and games - SI3D ’08, ACM Press, p. 183.

[15] Williams, L. Casting curved shadows on curved surfaces. ACM SIGGRAPH

Computer Graphics 12, 3 (Aug. 1978), 270–274.

[16] Williams, L. Pyramidal parametrics. ACM SIGGRAPH Computer Graphics

(37)

Appendix A

Supplemental Information

A.1 Questionnaire that was presented during the

experiment.

Following questions was asked:

• Was there anything salient or dissimilar within the shadows that you noticed? (not mandatory).

• Do you have any thoughts about the experiment? If you do, what was it? (not mandatory).

A.2 Answers from questionnaire.

1. Was there anything salient or dissimilar within the shadows that you noticed? (not mandatory)

• "The one on the right were a lot straighter and the one on the left was blurry but had a darker shade."

• "On one picture there was a low texture resolution where the shadows were. But say it only on one of the pictures."

• "I perceived the textures to be sharper in some images and more faded in other."

• "Think i noticed some diﬀerences between some images but im not com-pletely sure whether there was an actual diﬀerence or if i saw it because i was looking for something and tricked myself."

• "Shadows was a bit lower res on one side." • "No."

• "None"

• "I barely noticed any diﬀerence at all. In one of the choices I might have seen something diﬀerent where the shadow was not as sharp. It looked sort of matted out in the edges. That was nice."

• "On one picture there was a very blurry texture where the shadows were." 27

(38)

28 Appendix A. Supplemental Information

2. Do you have any thoughts about the experiment? If you do, what was it? (not mandatory).

• "It would have been nice if it wasn’t fading in and out that fast." • "The fading gave me a little eye strain."

• "On some of the comparisons i did not notice a diﬀerence."

• "It was interesting, but I did not know what to look for in the images when I only got one second to compare two images."

• "Similar images after each other made the response of pressing down the key hard to notice."

• "It was a bit annoying with the fading images."

• "I liked the way the images was presented. Only having a short time to view made me look closer."

(39)

A.2. Answers from questionnaire. 29 6KDGRZWH[WXUH'LVFUHSDQF\7HVW&RQVHQW)RUP 7KDQN\RXIRUFRPLQJKHUHDQGVKRZLQJLQWHUHVWLQWKLVWHVW 0\QDPHLV-RQDV%HUJJUHQDQG,¶PDWHFKQLFDODUWLVWVWXGHQWDW%7+XQGHUWKHVXSHUYLVLRQRI 3HWDU-HUþLü:HDUHDVNLQJ\RXWREHLQDVWXG\WRKHOSXVDSSUDLVHKXPDQH\H¶VDELOLW\WRQRWLFH GLVSDULWLHVDQGWRDGYDQFHWKHXVDJHRIVKDGRZPDSSLQJIRUDEDFKHORU¶VGHJUHHSURMHFW ,QWKLVWHVW Ɣ <RXZLOOEHDVNHGWRJRWKURXJKDVHULHVRISLFWXUHVRIGLIIHUHQWHQYLURQPHQWVDQG FKRRVHZKLFKRQH\RXSUHIHUPRUH Ɣ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þLü 7HOHSKRQHQXPEHU (PDLOSMU#EWKVH (PDLO-REQ#VWXGHQWEWKVH

(40)

30 Appendix A. Supplemental Information

Experiment Description

Experiment overview: By participating in this experimentation, you will be assigned the task to observe several different side-by-side images of environments on a computer and choose which one you prefer the most by pressing specific keys on the keyboard.

This experiment will take 10 – 20 minutes.

Test information:

● You’ll watch through fast paced side-by-side images and choose which one you find the most satisfactory.

● The images will be visible for a few seconds before becoming non-visible for a time before becoming visible again, this will repeat until participant chooses the most preferable image by pressing left-key for the left image or right-key for the right image.

● After completing the test, you will be handed an improvement questionnaire paper about the test to fill out. You can choose to leave it blank, but if you have an opinion of the test, feel free to write it down.

After fulfilling these points your task will be considered completed. Thank you for your cooperation.

(41)

(42)

Evaluating user preference when applying mipmap LOD in shadow covered textures