3D Graphics Rendering

2026

• Uniform Evaluation Metrics

  Last modified: January 13, 2026  |  Open article page

  The following (work in progress…)

  The following metrics are full-reference metics like the conventional PSNR and SSIM.

  • FLIP
  • LPIPS: git
  • pu-PSNR
  • ColorVideoVDP: Here is the Project Page and git repo.
  • Intel’s CGVQM

  Black Box Approaches

  1. NVidia’s FLIP

  FLIP is an excellent tool for visualizing and communicating errors in the rendered images, both for low dynamic range (LDR) and high dynamic range (HDR). Big thanks to the NVlabs for making the tool publicly available with the source code. The FLIP is a command line interface (CLI) tool. I am on Windows platform:

  1. First, git clone --recursive https://github.com/NVlabs/flip.git, then cd flip
  2. BUILD:

     mkdir build
     cd build
     cmake ..
     cmake --build .
     

     After this, the flip.exe should be under build/Debug. Similar way this can be extended to Debug and Release version. As per discussion, Release mode performs better

  // for Release
  mkdir Release
  cd Release
  cmake -DCMAKE_BUILD_TYPE=Release ..
  
  // for Debug
  mkdir Debug
  cd Debug
  cmake -DCMAKE_BUILD_TYPE=Debug ..
  

  1. RUNNING

     // go to Release/Debug folder and run
     .\flip.exe -r .\reference.png -t .\test.png
     // similarly
     .\flip.exe -r .\reference.exr -t .\test.exr
     

     Visually, color represents the error map. Black means no perceived error, and yellow represents high error. For more details, see the Technical Blog and the Flip Publication.

  There is an UI version of FLIP implementation, named as FLOP. Find in git with the blog explaining every step clearly. However, I guess the repository is not well-updated (2025) and there are several bugs that need to be resolved.

  ---

2025

• Perceptual Evaluation Metrics

  Last modified: November 05, 2025  |  Open article page

  (work in progress…)

  The evaluation of 3D rendering content under immersive experiences, i.e., extended reality (XR), is challenging and primarily user study-dependent. However, the subjective assessment is time-consuming, expensive, and user-biased. On the contrary, the objective evaluation metrics are fast, mathematically accurate, and more reliable. Nonetheless, the conventional pixel-based approaches, e.g., MSE, PSNR, and SSIM, fail to assess the true human evaluation process. Conversely, the perceptual evaluation metrics under the Visual Difference Predictor (VDP) genre are the most accurate evaluation strategy.

  Foveation-Considered Metrics

  VDP (HDR-VDP) genere metrics also referred to the White Box Approaches.

  However, the current popular VDP models have several limitations. For instance, HDR-VDP-2 is limited to achromatic and static; FovVideoVDP considers spatio-temporal with foveation; however limited to the achromatic; and ColorVideoVDP considers spatio-temporal, chromatic, XR-ready, differentiable; nonetheless, still lack explicit binocular 3D modeling and comfort prediction.

  FovVideoVDP

  The FovVideoVDP is a video/ image difference metric that models the spatial, temporal, and peripheral aspects of perception (See project page, git repo). It is a full reference metrics and to the best of my knowledge, this is the only metric that considers eccentricity and foveated aspect of vision. Unlike the other metrics, e.g., PSNR, SSIM, FovVideoVDP requires the physical display data (e.g., standard_4k, standard_fhd, viewing conditions (size, resolution, peak luminance, viewing distance, etc.). It works with both SDR and HDR content.

  The FovVideoVDP reports a quality score in JOD (Just-Objectionable-Difference) units: 10 = no visible difference, lower is worse (can be < 0 for very strong/unrelated differences).

  Use

  Open terminal (e.g., Windows terminal or, Visual Studio Code or anyother) and install using pip install torch pyexr pyfvvdp or simply pip install pyfvvdp should also work fine. Afterwards, you can run:

  fvvdp --ref .\sponza_uni_0_32.png --test .\sponza_fov_0.png --display standard_4k --foveated --heatmap supra-threshold --output-dir out/
  fvvdp --ref .\sponza_uni_0_32.png --test .\sponza_uni_0.png --display standard_4k --heatmap supra-threshold --output-dir out/ 
  

  For instance, here is a heatmap comparison between Crytek Sponza scene generated by uniform (constant 4 samples-per-pixel) path tracing (FovVideoVDP=3.8321 [JOD]) and foveated (maximum 4 samples-per-pixel and minimum 1 sample-per-pixel with different degradation strategies) path tracing (FovVideoVDP=3.4431 [JOD]) with respect to the uniform (32 samples-per-pixel) path tracing. For foveated evaluation, use the --foveated command. Figure 6 is the FLIP evaluation (Mean: 0.201726, Weighted median: 0.263627, 1st weighted quartile: 0.182165, 3rd weighted quartile: 0.353026, Min: 0.001038, Max: 0.703269) between the uniform (4 samples-per-pixel) and foveated (max 4 samples-per-pixel) heatmaps (uniform, fig 4 and foveated, fig 5) generated by FovVideoVDP.

  Figure 1. Foveated Path Tracing Evaluation.

  p.s., FovVideoVDP also can perform batch evaluation, for instance fvvdp --ref "ref/*.png" --test "test/*.png" --display standard_fhd --heatmap supra-threshold --output-dir out/

• Benchmarking and Profiling

  Last modified: November 05, 2025  |  Open article page

  (work in progress…)

  In real-time rendering, often the frame rate (fps) is presented as the computational improvement. However, the frame rate sometimes miss the important evaluation criteria. Therefore, the frame time (aka. render time (in millisecond)) could be a better option. For more, see this excellent article.

  Many of the existing graphics debugger tools work in this philosophy to evaluate per frame rendering time to find the drawbacks. Some of the popular debuggers are:

  • NVidia: NVIDIA Nsight Graphics, NSight Developer Tools, NSight Visual Studio add-ons
  • Microsoft:
    • RenderDoc, RenderDoc UE Plugin (Vulkan, D3D11, D3D12, D3DR, OpenGL, OpenGL ES supported)
    • PIX DX12, support DX12
    • Visual Studio Graphics Analyzer on Windows, Linux, Android, and Nintendo Switch
  • Intel: VTune, Graphics Performance Analyzer (GAP)
  • AMD: AMD GPU PerfStudio, Radeon Developer Tools Suite, Radeon Ray Tracing Analyzer

  Other than, there are several other debuggers, e.g.,

  • awesome-benchmark
  • TauBench: A Dynamic Benchmark for Graphics Rendering (Dataset Reference Frames from Tampere University’s VGA Group
  • qv: histogram, pixel value inspection, and statistics of rendered image
  • NVAPI and Shader Execution Recording
• Statistics For Rendering Engineers

  Last modified: October 29, 2025  |  Open article page

  Some Statistical Analysis Tools

  • G*Power often used to determine the sample size (aka. number of participants) based on the statistical method/s will be used
  • Two tools I find very handy, i.e., jamovi and jasp for do the analysis without writing any additional line of code
  • Many researchers rely on the statistical software like IBM SPSS or R-Language
  • Other than those, the Python is the most powerful for data analysis. Python there are numerous libraries, e.g., NumPy, SciPy, Pandas are frequently used. Additionally, someone could check Python pingouin for statistical analysis. Someone can use pandas for data manipulation, SciPy or statsmodels for tests, and matplotlib or seaborn for plotting results. Moreover, If someone wants to work on eye-tracking data analysis, the package, pymovements is very handy.
• Computer Graphics Resources

  Last modified: October 20, 2025  |  Open article page

  3D Model For Research

  Here you can find some 3D scenes and models frequently used in the research papers.

  1. ORCA: Open Research Content Archive
  2. McGuire Computer Graphics Archive 👍🏼
  3. Benedikt Bitterli’s Rendering Resources 👍🏼
  4. Stanford 3D Scanning Repository
  5. Clara.io
  6. Frequently used small project scenes: Benedikt Bitterli’s Rendering Resources 👍🏼 👍🏽 (Falcor: download)
  7. advanced: Moana Island Scene: Disney
  8. HDRI light texture
  9. polygon heaven
  10. hdri heaven
  11. From GitHub
  12. Commonly used Scenes and Assets
  13. few good 3D models
  14. ray tracing references
  15. pbr database 👍🏽
  16. pbr-based databased for CG artists

2022

• Objects and Materials Part 1 (Wavefront 3D Model Specification)

  Last modified: June 24, 2022  |  Open article page

  1.1. wavefront .obj

  the OBJ format, the MTL format was used and documented by Wavefront Technologies in the 1990s in association with its Advanced Visualizer software. Object files can be in ASCII format (.obj) or binary format (.mod). Vertices are stored in a counter-clockwise, and in object space (un-normalized).

  1. v : vertex
     • find in different ranges, e.g., 0.0-1.0,
     • v 1380.6033 -115.9791 -782.5203
     • v -0.122936 1.00032 0.35065
     • (x, y, z [,w]) w is optional and defaults to 1.0
  2. vt : texture (uv) position [-1.0 to 1.0] of each texture coordinate vertex
     • vt 0.5767 0.5096 0.0000 (sponza scene)
     • List of texture coordinates, in (u, [,v ,w]) coordinates, ~these will vary between -1 and 1. v, w are optional and default to 0.~
  3. vn: vertex normal, in (x,y,z) form
     • vn -0.1776 -0.8781 -0.4443 (sponza scene)
  4. f: polygon faces that make each polygon defined as a list of vertices. vertex/texture/normal
     • f 147019/91297/142810 147015/91295/142806 147014/91131/142728 147022/91139/142813 (sponza scene)
  5. s: smoothing group
  6. vp -> Parameter space vertices in ( u [,v] [,w] )
     • Free-form curve/surface attributes
  7. l # Line element (see below)

  1.2. mtl (material template library)

  he MTL specification, as published in 1995, includes a substantial number of options that are typically not supported in today’s software.

  // sponza scene material (tail)
  newmtl roof                                    // sequence of material definitions, each beginning with the keyword
  
  	Ns 10.0000                             // specular highlights, normally 0-1000, higher value resulting in a tight,
                                                 // concentrated highlight. Default is 0.0
  	Ni 1.5000                              // optical density (aka index of refraction), 0.001-10. A value of 1.0 means that light
                                                 // does not bend as it passes through an object
  
  // alpha values (d = 1 means fully opaque, Tr = 0 means fully opaque  
  	d 1.0000                               // alpha value (aka. factor for dissolve, how much this material dissolves into the background. A factor of 1.0
                                                 // is fully opaque. A factor of 0.0 is completely transparent
  	Tr 0.0000                              // alpha, defines the transparency of the material to be alpha.
                                                 // The default is 0.0 (not transparent at all).
                                                 // The quantities d and Tr are the opposites of each other, and specifying transparency or
                                                 // nontransparency is simply a matter of user convenience.
  
  //
  	Tf 1.0000 1.0000 1.0000                // transmission filter
  	illum 2                                // illumination model
                                                 // illum = 1 indicates a flat material with no specular highlights, so the value of Ks is not used.                                          // illum = 2 denotes the presence of specular highlights, and so a specification for Ks is required.
  	Ka 1 1 1                               // ambient color 0-1 (phong reflection model/ phong illumination/ phong lighting)
  	Kd 1 1 1                               // diffuse color (0-1)
  	Ks 0.0000 0.0000 0.0000                // specular color
  
  	Ke 0.0000 0.0000 0.0000                // emissive coeficient. It goes together with ambient, diffuse and specular and represents the
                                                 // amount of light emitted by the material.
  	map_Ka textures\sponza_roof_diff.png   // ambient color texture file  
  	map_Kd textures\sponza_roof_diff.png   // Color texture file to be applied to the diffuse reflectivity of the material.
                                                 // During rendering, map_Kd values are multiplied by the Kd values to derive the RGB components
  

  illum

  illum is interesting, there are a number of options, but as far as Wave is concerned, there are only two.

   Illumination    Properties that are turned on in the
   model           Property Editor
  
   0		Color on and Ambient off
   1		Color on and Ambient on
   2		Highlight on
   3		Reflection on and Ray trace on
   4		Transparency: Glass on
   		Reflection: Ray trace on
   5		Reflection: Fresnel on and Ray trace on
   6		Transparency: Refraction on
   		Reflection: Fresnel off and Ray trace on
   7		Transparency: Refraction on
   		Reflection: Fresnel on and Ray trace on
   8		Reflection on and Ray trace off
   9		Transparency: Glass on
   		Reflection: Ray trace off
   10		Casts shadows onto invisible surfaces
  

  1.3. textures

  • JPG and PNG are known to be supported.

  References

  • paulbourke.net
  • paulbourke material