Visual Generative AI for Extended Reality: A Survey. Release 23 October 2026.
31 selected methods and systems, eight dimensions, 248 categorical evidence entries. E/ identifies a detailed evidence record, distinct from catalogue numbering. Each row below preserves the original cell note and source locator. Evidence categories are non-ordinal; a record can combine separately inspected configurations and does not certify a jointly validated system.
E targeted evaluation (including mixed/negative results); D implementation/demonstration; C source claim; I analytical inference; NR no evidence located in inspected material; NA outside the selected output scope. Source and configuration limits must be read with the code.
Catalogue crosswalk · Cell data · Inspection records · Search trace · Numerical source editions
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | D | Multiview shape, normal maps, and relighting are demonstrated; CLIP retrieval scores are not treated as spatial accuracy. | https://arxiv.org/html/2209.14988 §4 |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2209.14988 §3.2–5 |
| temporal | NA | NA for temporal evolution of the core static asset. Rendered viewpoints are indexed under spatial evidence; the inspected core method does not generate object dynamics or implement a retained-modification process. | https://arxiv.org/html/2209.14988 §3.2–5 |
| control | E | Text matching is evaluated using CLIP R-Precision; this does not assess precise spatial constraints. | https://arxiv.org/html/2209.14988 §4 Table 1 |
| interaction | NR | No runtime action-interaction test was located in the inspected sections. | https://arxiv.org/html/2209.14988 §3.2–5 |
| export | D | The official website provides marching-cubes mesh export. | https://dreamfusion3d.github.io/ Mesh exports |
| runtime | E | The source reports a compute-time decomposition, divided evenly between NeRF rendering and diffusion-model evaluation, for 15,000 optimization iterations on four TPUv4 chips (about 1.5 hours). This evaluates optimization cost, not runtime display frame rate. | https://arxiv.org/html/2209.14988v1 Section 3.2, item 4 (Optimization) |
| xr | NR | No headset deployment or XR user experiment was located in the inspected sections. | https://arxiv.org/html/2209.14988 §3.2–5 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | D | Shape and texture comparisons are demonstrated; this table does not reinterpret overall visual preference as geometric accuracy. | https://arxiv.org/html/2211.10440 §5 Figures 3–5 |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2211.10440 §4–6 |
| temporal | NA | NA for temporal evolution of the core static asset. Rendered viewpoints are indexed under spatial evidence; the inspected core method does not generate object dynamics or implement a retained-modification process. | https://arxiv.org/html/2211.10440 §4–6 |
| control | D | Editing includes re-optimization after prompt changes and image conditioning. | https://research.nvidia.com/labs/cosmos-lab/magic3d/ Editing Capabilities |
| interaction | NR | No runtime action-interaction test was located in the inspected sections. | https://arxiv.org/html/2211.10440 §4–6 |
| export | D | The official page provides textured-mesh downloads. | https://research.nvidia.com/labs/cosmos-lab/magic3d/ High-Resolution 3D Meshes |
| runtime | E | Coarse and fine optimization total 40 minutes on 8 A100 GPUs; this is generation time, not display FPS. | https://arxiv.org/html/2211.10440 §5 Speed evaluation |
| xr | NR | No headset deployment or XR user experiment was located in the inspected sections. | https://arxiv.org/html/2211.10440 §4–6 |
TRELLIS quantitative values in Table 3A use arXiv:2412.01506v1. The formal bibliography cites the CVPR publication.
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Representation reconstruction evaluates geometry using CD, F-score, and normal-map error; the evidence is limited to the reconstruction protocol. | https://arxiv.org/html/2412.01506v1 §4.1 |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2412.01506v1 §3.2–4.4 |
| temporal | NA | NA for temporal evolution of the core static asset. Rendered viewpoints are indexed under spatial evidence; the inspected core method does not generate object dynamics or implement a retained-modification process. | https://arxiv.org/html/2412.01506v1 §3.2–4.4 |
| control | D | Region deletion, addition, replacement, and asset variations are demonstrated. | https://arxiv.org/html/2412.01506v1 §4.4 Figure 7 |
| interaction | NR | No runtime action-interaction test was located in the inspected sections. | https://arxiv.org/html/2412.01506v1 §3.2–4.4 |
| export | D | The official implementation documents textured GLB conversion and Gaussian PLY export. GLB conversion includes mesh postprocessing, UV parameterization and texture baking. Implementation inspected; model/export not executed and engine or headset validation not inferred. | https://github.com/microsoft/TRELLIS/blob/442aa1e1afb9014e80681d3bf604e8d728a86ee7… README minimal example, lines 164–176; trellis/utils/postprocessing_utils.py, to_glb, lines 399–453 |
| runtime | NR | No verifiable generation, rendering, or interaction timings were located in the inspected sections. | https://arxiv.org/html/2412.01506v1 §3.2–4.4 |
| xr | NR | No headset deployment or XR user experiment was located in the inspected sections. | https://arxiv.org/html/2412.01506v1 §3.2–4.4 |
The temporal D concerns Hunyuan3D Studio skeleton, skinning and retargeting tools, not motion generated by the shape model.
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | ShapeVAE reconstruction is measured using volume/surface IoU; generated shapes are assessed using cross-modal similarity. | https://arxiv.org/html/2501.12202v1 §5.1 Tables 1–2 |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2501.12202v1 §2,4,5,6 |
| temporal | D | Studio demonstrates animation using skeletons, skinning, and motion retargeting; this is not native 4D generation by the generator. | https://arxiv.org/html/2501.12202v1 §6.3 |
| control | E | Image-condition matching is assessed using automatic metrics and user evaluation. | https://arxiv.org/html/2501.12202v1 §5.3 |
| interaction | NR | No runtime action-interaction test was located in the inspected sections. | https://arxiv.org/html/2501.12202v1 §2,4,5,6 |
| export | D | The official application implements mesh export and download in GLB, OBJ, PLY and STL, with separate textured and untextured paths. Implementation inspected; model/export not executed and engine or headset validation not inferred. | https://github.com/Tencent-Hunyuan/Hunyuan3D-2/blob/f8db63096c8282cb27354314d896fe… gradio_app.py: export_mesh, lines 85–94; Export UI, lines 460–471; on_export_click, lines 599–638; SUPPORTED_FORMATS, line 690 |
| runtime | NR | No verifiable generation, rendering, or interaction timings were located in the inspected sections. | https://arxiv.org/html/2501.12202v1 §2,4,5,6 |
| xr | NR | No headset deployment or XR user experiment was located in the inspected sections. | https://arxiv.org/html/2501.12202v1 §2,4,5,6 |
The cost E concerns the measured decoder experiment. Approximate full-generation timings are source claims and do not inherit E.
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Reconstruction is evaluated using Mesh Distance, CD, F-score, and normal error; generation is assessed using alignment and human evaluation. | https://arxiv.org/html/2512.14692v1 §4.1–4.2 |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2512.14692v1 §1,3,4; Appendix A,D |
| temporal | NA | NA for temporal evolution of the core static asset. Rendered viewpoints are indexed under spatial evidence; the inspected core method does not generate object dynamics or implement a retained-modification process. | https://arxiv.org/html/2512.14692v1 §1,3,4; Appendix A,D |
| control | D | PBR materials can be generated independently given geometry and a reference image. | https://arxiv.org/html/2512.14692v1 §4.3 |
| interaction | NR | No runtime action-interaction test was located in the inspected sections. | https://arxiv.org/html/2512.14692v1 §1,3,4; Appendix A,D |
| export | D | The official example exports generated geometry and PBR attributes to GLB through O-Voxel postprocessing. Transparency requires activation after import because the default export mode is OPAQUE. Implementation inspected; model/export not executed and engine or headset validation not inferred. | https://github.com/microsoft/TRELLIS.2/blob/75fbf0183001ed9876c8dbb35de6b68552ee08… README: Export to GLB, lines 151–174; o-voxel/README.md: Export to GLB, lines 110–127 |
| runtime | E | Decoder runtime is measured in reconstruction experiments: 0.077/0.301 s at 512³/1024³ on one A100. Separately, the introduction reports 3/17/60 s asset generation at 512³/1024³/1536³ on one H100; these are author-reported generation figures. Neither measures display latency. | https://arxiv.org/html/2512.14692v1 Section 4 implementation details; Section 4.1, Table 1, Dec. (s); Section 1 generation claim |
| xr | NR | No headset deployment or XR user experiment was located in the inspected sections. | https://arxiv.org/html/2512.14692v1 §1,3,4; Appendix A,D |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Shape CD/F-score and part-overlap IoU receive targeted evaluation. | https://arxiv.org/html/2506.05573 §4 Evaluation Metrics |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2506.05573 §3.4–4.2 |
| temporal | NA | NA for temporal evolution of the core static asset. Rendered viewpoints are indexed under spatial evidence; the inspected core method does not generate object dynamics or implement a retained-modification process. | https://arxiv.org/html/2506.05573 §3.4–4.2 |
| control | D | The part-count setting controls the granularity of the output decomposition. | https://arxiv.org/html/2506.05573 §3.4, §4 |
| interaction | I | The possibility of transforming independent meshes separately is an inference from the representation; joint-based or physical interaction has not been verified. | https://arxiv.org/html/2506.05573 §4 Evaluation Metrics |
| export | D | The official inference script exports each generated part as a separate GLB, a merged object GLB, and a part-to-file manifest. Implementation inspected; model/export not executed and engine or headset validation not inferred. | https://github.com/wgsxm/PartCrafter/blob/3d773bf02fad51c7ab31a5615573fec93b287b30… scripts/inference_partcrafter.py: per-part and composite export and manifest, lines 159–181; scripts/inference_partcrafter_scene.py, lines 143–165 |
| runtime | E | A 4-part object takes 34 seconds on average on a single H20; this is generation time. | https://arxiv.org/html/2506.05573 §4 Evaluation Metrics, Table 1 |
| xr | NR | No headset deployment or XR user experiment was located in the inspected sections. | https://arxiv.org/html/2506.05573 §3.4–4.2 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Section 4.2 includes user ratings of 3D structure/completeness; these are not errors against real geometry. | https://arxiv.org/html/2303.11989 §4.2, Table 1 |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2303.11989 §4.2,4.4–4.5,5 |
| temporal | NA | NA for temporal content or state generation in the core static-mesh output. Construction trajectories address spatial completion; they are not a separately evaluated temporal-continuity or retained-modification protocol. | https://arxiv.org/html/2303.11989 §4.2,4.4–4.5,5 |
| control | E | CLIP scores on rendered scene images evaluate text-image alignment; precise layout control remains limited and is not established by this proxy. | https://arxiv.org/html/2303.11989v2 Section 4 (Evaluation Metrics), Section 4.2, Table 1; Sections 4.4-4.5 |
| interaction | D | The official webpage supports mouse-based mesh exploration; this evidence is limited to camera navigation, not object manipulation. | https://lukashoel.github.io/text-to-room/ Interactive 3D Mesh Viewer |
| export | D | Textured meshes are produced, and the official project provides downloads of complete scene meshes. | https://lukashoel.github.io/text-to-room/ Download meshes; Interactive 3D Mesh Viewer |
| runtime | C | The implementation description reports about 50 minutes per scene on one RTX 3090, with 20 trajectories and 10 frames each. A dedicated timing comparison, profile, or repeated measurement was not located; the value is retained as an author-reported generation estimate. | https://arxiv.org/html/2303.11989v2 Section 4 (Implementation Details) |
| xr | C | The authors identify AR/VR content creation as an application; no headset experiment was located in the inspected sections. | https://arxiv.org/html/2303.11989 §1 Introduction |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Camera error, reprojection/depth error, and the SfM reconstruction ratio are evaluated. | https://arxiv.org/html/2302.01133 §4.1–4.3 |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2302.01133 §4.1–4.3,5; Appendix A |
| temporal | E | Decoder-optimization ablations and reprojection error address inter-frame consistency under camera motion, not dynamic objects. | https://arxiv.org/html/2302.01133 §4.2–4.3 |
| control | E | Recovery error for the prescribed camera motion receives targeted evaluation; text is assessed separately using CLIP. | https://arxiv.org/html/2302.01133 §4.1–4.3 |
| interaction | NR | No runtime action-interaction test was located in the inspected sections. | https://arxiv.org/html/2302.01133 §4.1–4.3,5; Appendix A |
| export | I | A unified mesh and Poisson processing support a potential delivery route; project demonstrations do not constitute verified export. | https://scenescape.github.io/ 3D scene visualization |
| runtime | E | 50 frames take approximately 2.5 hours on a Tesla V100; the stage timings in the appendix cannot be treated as display FPS. | https://arxiv.org/html/2302.01133 Appendix A, Table 3 |
| xr | NR | No headset deployment or XR user experiment was located in the inspected sections. | https://arxiv.org/html/2302.01133 §4.1–4.3,5; Appendix A |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Depth error, camera error, and the SfM reconstruction ratio are evaluated, including validation with a 600-frame circular trajectory. | https://arxiv.org/html/2302.01330 §V-B,V-E |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2302.01330 §V-B,V-E,V-F,VI |
| temporal | E | Camera-path view consistency is evaluated through camera recovery/SfM and user judgments of rendered trajectories; Section V-E also reconstructs a 600-frame circular traversal. This indexes coherence across views of a static scene, not object dynamics or retention of a user modification. The spatial and temporal entries refer to overlapping observations, not independent tests. | https://arxiv.org/html/2302.01330v3 Section V-B (CE and SfM rate); Section V-E (User Study and View Consistency), Figures 15-16 |
| control | D | Scene/style latent variables can be varied, and rendering follows a prescribed camera path; these are not precise layout constraints. | https://scene-dreamer.github.io/ Framework; style interpolation |
| interaction | NR | Official demonstrations use predefined camera paths; online interaction driven by user input was not located in the inspected material. | https://scene-dreamer.github.io/ Unbounded 3D scene generation demos |
| export | NR | No export pipeline was located in the inspected sections. | https://arxiv.org/html/2302.01330 §V-B,V-E,V-F,VI |
| runtime | C | Section IV-C reports approximately 2 seconds to render one 960 x 540 frame on an A100 GPU. This is a standalone rendering estimate; no corresponding timing comparison, profile, or repeated-measurement protocol was located in that passage. It is not generation time or input-to-display latency. | https://arxiv.org/html/2302.01330v3 Section IV-C (Inference), High-resolution renderings |
| xr | NR | No headset deployment or XR user experiment was located in the inspected sections. | https://arxiv.org/html/2302.01330 §V-B,V-E,V-F,VI |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Human layout evaluation considers spatial boundaries, collisions, reachability, and realism; these are not geometric reconstruction errors. | https://arxiv.org/html/2312.09067v2 §4.3 |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2312.09067v2 §3–5; Appendix A.1 |
| temporal | NA | NA for temporal content or retained-state generation in the core scene-assembly method. The separate simulator-navigation evaluation concerns agent interaction; it does not evaluate persistence of a user modification. | https://arxiv.org/html/2312.09067v2 §3–5; Appendix A.1 |
| control | E | Text adherence is evaluated using CLIP, human evaluation, and spatial-constraint ablations. | https://arxiv.org/html/2312.09067v2 §4.1–4.3 |
| interaction | E | NoveltyTHOR agent navigation is evaluated using Success/SPL; this does not establish headset use or object manipulation. | https://arxiv.org/html/2312.09067v2 §5, Table 2 |
| export | D | Meshes, visibility points, and colliders imported into AI2-THOR form a simulator scene. | https://arxiv.org/html/2312.09067v2 §3 |
| runtime | C | The source reports about three minutes to generate one room using a MacBook M1, including GPT-4 API calls and layout optimization. A timing comparison, profile, or repeated measurement was not located; this remains an author-reported implementation estimate. | https://arxiv.org/html/2312.09067v2 Appendix A.1 (Efficiency and Cost) |
| xr | NR | No headset deployment or XR user experiment was located in the inspected sections. | https://arxiv.org/html/2312.09067v2 §3–5; Appendix A.1 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | D | Hierarchical 3D geometry and explorable results are demonstrated; the inspected quantitative tables mainly report 2D image metrics. | https://arxiv.org/html/2507.21809v1 §2.4,3.2–3.4 |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2507.21809v1 §2.4–2.6; §3.2–3.4 |
| temporal | NA | NA for temporal content or retained-state generation in the core static-scene output. Demonstrated object transforms are indexed as interaction; a path-continuity or modification-retention protocol was not located in the inspected material. | https://arxiv.org/html/2507.21809v1 §2.4–2.6; §3.2–3.4 |
| control | E | Text/image alignment is evaluated using CLIP-related metrics; this is not equivalent to precise 3D layout control. | https://arxiv.org/html/2507.21809v1 §3.2–3.3, Tables 3–4 |
| interaction | D | Translation, rotation, and scaling of independent objects are demonstrated; targeted usability or latency evaluation was not located. | https://arxiv.org/html/2507.21809v1 §3.4 |
| export | D | The text describes mesh formats and Unity/Unreal integration pipelines. | https://arxiv.org/html/2507.21809v1 §3.4 |
| runtime | NR | TensorRT, caching, multi-GPU operation, and other optimizations are listed; a verifiable timing/FPS protocol was not located in the inspected sections. | https://arxiv.org/html/2507.21809v1 §2.6 |
| xr | C | The authors state compatibility with Vision Pro/Quest; a device-evaluation protocol was not located within the inspected scope. | https://arxiv.org/html/2507.21809v1 §3.4 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Table 5 evaluates single-image generative reconstruction using MVS pseudo-ground truth and Precision/Recall/F1/AUC; this is not geometric ground truth for all world generation. | https://arxiv.org/html/2604.14268v1 §8.1.3, Table 5 |
| stereo | NR | No evaluation of paired-eye consistency or binocular viewing in a headset was located in the inspected sections. | https://arxiv.org/html/2604.14268v1 §1–2,7.2,8.1.3,8.1.5 |
| temporal | D | Keyframe memory supports continuous camera-based expansion; this concerns preservation of a static world, not the evolution of dynamic objects. | https://arxiv.org/html/2604.14268v1 §2 overview; §5 WorldStereo 2.0 |
| control | E | WorldStereo 2.0 camera trajectories receive targeted evaluation using RotErr/TransErr/ATE; the evidence is limited to the component protocol. | https://arxiv.org/html/2604.14268v1 §8.1.3, Table 6 |
| interaction | D | Virtual-agent navigation and stair collision-proxy demonstrations are provided; input-to-feedback latency measurement was not located. | https://arxiv.org/html/2604.14268v1 §8.1.5, Figure 22 |
| export | I | GS and collision-proxy meshes are present; general file export or an external-engine interface was not verified within the inspected scope. | https://arxiv.org/html/2604.14268v1 §7.2 Mesh Extraction; §8.1.5 |
| runtime | E | Table 10 reports offline generation times by stage on H20 hardware; the GPU count is unconfirmed, and these are not rendering FPS. | https://arxiv.org/html/2604.14268v1 §1 offline scope; §8.1.5, Table 10 |
| xr | C | VR is listed as an application direction; no headset-device or user experiment was located in the inspected sections. | https://arxiv.org/html/2604.14268v1 §1; §8.1.5 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Human evaluation of 3D structure; not validation of metric geometry. | https://arxiv.org/html/2311.17984v2 §4/Table 2 |
| stereo | NR | No binocular/IPD evaluation located in the inspected material. | https://arxiv.org/html/2311.17984v2 §3–5; Supplement S1–S2 |
| temporal | E | Paired motion-quality judgments by 26 evaluators; preference for 4D-fy against MAV3D was 41%, without a significant difference for this criterion. This does not test long-term persistence. | https://arxiv.org/html/2311.17984v2 Section 4.1 User study; Table 2; Supplement S2 |
| control | E | Text alignment is evaluated with CLIP scores on rendered video frames and paired human judgments; this does not test online action control. | https://arxiv.org/html/2311.17984v2 Section 4.1, Table 2; Supplement S2 (Text Alignment) |
| interaction | NR | No closed-loop interaction located in the inspected material. | https://arxiv.org/html/2311.17984v2 §3–5; Supplement S1–S2 |
| export | NR | No usable 3D asset export located in the inspected material. | https://arxiv.org/html/2311.17984v2 §3–5; Supplement S1–S2 |
| runtime | E | Supplement S1 compares 71 ms per rendered frame with 68 ms for a static MVDream-based model. Separately, optimization is reported as 2+2+19 hours on an A100 with about 80 GB VRAM. E refers to the rendering-time comparison; neither value establishes input-to-display latency. | https://arxiv.org/html/2311.17984v2 Supplement S1 (Computation; Runtime) |
| xr | NR | No headset or XR system validation located in the inspected material. | https://arxiv.org/html/2311.17984v2 §3–5; Supplement S1–S2 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Video-to-4D evaluated using ground-truth novel views. | https://arxiv.org/html/2312.17142v3 §4.1/Table 3 |
| stereo | NR | No binocular/IPD evaluation located in the inspected material. | https://arxiv.org/html/2312.17142v3 §3–5, Appendix A–D |
| temporal | E | Video FVD and human evaluation of motion quality; consistency after refinement is assessed mainly qualitatively. | https://arxiv.org/html/2312.17142v3 §4.1/Table 3; Fig.14 |
| control | D | Selecting a driving video changes the motion. | https://arxiv.org/html/2312.17142v3 §4.3/Fig.15 |
| interaction | NR | No closed-loop interaction located in the inspected material. | https://arxiv.org/html/2312.17142v3 §3–5, Appendix A–D |
| export | D | Textured mesh sequences composed and rendered in Blender. | https://arxiv.org/html/2312.17142v3 §4.3/Fig.1 |
| runtime | E | A100: image-to-4D optimization takes 6.5 minutes, or 10 minutes including refinement. | https://arxiv.org/html/2312.17142v3 Table 1; §4 Implementation |
| xr | C | The authors list MR applications; headset validation not located. | https://arxiv.org/html/2312.17142v3 §5 Broader Impacts |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Novel-view PSNR/SSIM/LPIPS evaluate 3D reconstruction. | https://arxiv.org/html/2411.04928v1 §4.3/Table 1 |
| stereo | NR | No binocular/IPD evaluation located in the inspected material. | https://arxiv.org/html/2411.04928v1 §3–4 |
| temporal | E | VBench subject consistency; demonstrations of 4D novel views. | https://arxiv.org/html/2411.04928v1 §4.2/Table 2; §4.4 |
| control | D | Decoupled camera and temporal motion demonstrated; not closed-loop control. | https://arxiv.org/html/2411.04928v1 §3.2; §4.2/Fig.4 |
| interaction | NR | No closed-loop interaction located in the inspected material. | https://arxiv.org/html/2411.04928v1 §3–4 |
| export | NR | GS reconstructed; no deliverable export or engine integration located in the inspected material. | https://arxiv.org/html/2411.04928v1 §3.3–3.4 |
| runtime | NR | No runtime/latency evaluation located in the inspected material. | https://arxiv.org/html/2411.04928v1 §3–4 |
| xr | NR | No headset or XR system validation located in the inspected material. | https://arxiv.org/html/2411.04928v1 §3–4 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | 3DGS reconstruction and WorldScore 3D consistency. | https://arxiv.org/html/2506.04225v1 §5.2–5.3/Tables 2–3 |
| stereo | NR | No binocular/IPD evaluation located in the inspected material. | https://arxiv.org/html/2506.04225v1 §4–6, Appendix A |
| temporal | D | The world cache supports repeated generation over long horizons; dynamic-object persistence not located. | https://arxiv.org/html/2506.04225v1 §4.3 |
| control | E | WorldScore evaluation of camera control. | https://arxiv.org/html/2506.04225v1 §5.3/Table 3 |
| interaction | NR | No closed-loop interaction located in the inspected material. | https://arxiv.org/html/2506.04225v1 §4–6, Appendix A |
| export | NR | Point clouds used for subsequent 3DGS reconstruction; no deliverable asset export or engine integration located in the inspected sections. | https://arxiv.org/html/2506.04225v1 §5.2/Fig.5 |
| runtime | NR | No runtime/latency evaluation located in the inspected material. | https://arxiv.org/html/2506.04225v1 §4–6, Appendix A |
| xr | C | The abstract lists VR requirements; a headset implementation not located. | https://arxiv.org/html/2506.04225v1 Abstract; §6 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | NR | Object/scenario consistency measures are perceptual video proxies; the inspected evaluations do not establish multiview geometry or spatial consistency on revisiting. | https://arxiv.org/html/2508.13009v2 §5.1/Tables 1–2 |
| stereo | NR | No binocular/IPD evaluation located in the inspected material. | https://arxiv.org/html/2508.13009v2 §4–5 |
| temporal | E | Temporal consistency/motion smoothness evaluated on 597-frame sequences. | https://arxiv.org/html/2508.13009v2 §5.1–5.2 |
| control | E | Keyboard/mouse accuracy in Minecraft. | https://arxiv.org/html/2508.13009v2 Table 1 |
| interaction | D | Frame-level keyboard/mouse conditioning and streaming feedback; a closed loop with a headset has not been verified. | https://arxiv.org/html/2508.13009v2 §4.2 |
| export | NR | No usable 3D asset export located in the inspected material. | https://arxiv.org/html/2508.13009v2 §4–5 |
| runtime | E | Table 3 reports generation-throughput ablations of 15.49/21.03/25.15 FPS; Section 1 reports a single H100, and Section 5.1 specifies 352×640 output; these are not input-to-feedback latency measurements. | https://arxiv.org/html/2508.13009v2 §1; §5.1 Dataset; §5.3/Table 3; §6.1 |
| xr | NR | No headset or XR system validation located in the inspected material. | https://arxiv.org/html/2508.13009v2 §4–5 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | D | Demonstrations of scene revisiting and reappearance after occlusion. | https://arxiv.org/html/2604.08995v1 §5.2.2/Fig.11 |
| stereo | NR | No binocular/IPD evaluation located in the inspected material. | https://arxiv.org/html/2604.08995v1 §3–5 |
| temporal | D | Minute-scale memory demonstrated; no quantitative consistency table located. | https://arxiv.org/html/2604.08995v1 §5.2.1–5.2.2 |
| control | D | Qualitative demonstrations of character and camera control. | https://arxiv.org/html/2604.08995v1 §5.2.1/Figs.8–10 |
| interaction | D | Implemented action-conditioned streaming generation. | https://arxiv.org/html/2604.08995v1 §3.1–3.4 |
| export | NR | No usable 3D asset export located in the inspected material. | https://arxiv.org/html/2604.08995v1 §3–5 |
| runtime | E | Approximately 40 FPS at 720p; experiments use an asynchronous 8+1 GPU configuration. | https://arxiv.org/html/2604.08995v1 §5.2.3/Table 1 |
| xr | C | The introduction lists XR applications; headset validation not located. | https://arxiv.org/html/2604.08995v1 §1 |
The coded WorldPlay profile uses arXiv:2512.14614v2; numerical results in Table 5 use the PMLR volume 306 edition (sun26az).
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | MEt3R evaluation of revisited frames; a learned geometric-consistency proxy, not metric geometric ground truth. | https://arxiv.org/html/2512.14614v2 §4.1/Table 3 |
| stereo | NR | No binocular/IPD evaluation located in the inspected material. | https://arxiv.org/html/2512.14614v2 §3–5; Appendix C–E |
| temporal | E | Revisit consistency along round-trip trajectories; not a guarantee for unlimited duration. | https://arxiv.org/html/2512.14614v2 §4 Evaluation Protocol |
| control | E | Camera rotation/translation errors and action-representation ablations. | https://arxiv.org/html/2512.14614v2 §4/Table 5 |
| interaction | D | Online navigation; text-triggered event demonstrations. | https://arxiv.org/html/2512.14614v2 §3.5; §4.3/Fig.10 |
| export | NR | A separate reconstructor produces point clouds for demonstration; no deliverable asset export or engine integration located in the inspected sections. | https://arxiv.org/html/2512.14614v2 §4.3/Fig.9 |
| runtime | E | 720p generation at 24 FPS on 8×H800; throughput ablations, not input-to-display latency. | https://arxiv.org/html/2512.14614v2 §3.5; Table 4 |
| xr | NR | No headset or XR system validation located in the inspected material. | https://arxiv.org/html/2512.14614v2 §3–5; Appendix C–E |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | D | Environmental revisiting demonstrated; no metric-geometry evaluation. | https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/ Environmental consistency over a long horizon |
| stereo | NR | No binocular/IPD evaluation located in the inspected material. | https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/ Capabilities and Next steps |
| temporal | D | Minute-scale consistency demonstrated. | https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/ Environmental consistency over a long horizon |
| control | D | Examples of navigation and text-based weather changes/object addition. | https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/ Promptable world events |
| interaction | D | Public recordings of real-time interaction and SIMA agent examples. | https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/ Capabilities; Fueling embodied agent research |
| export | NR | No usable 3D asset export located in the inspected material. | https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/ Capabilities and Next steps |
| runtime | C | The authors report 720p/24 FPS; hardware and end-to-end latency are undisclosed. | https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/ Opening and Pushing the frontier |
| xr | NR | No headset or XR system validation located in the inspected material. | https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/ Capabilities and Next steps |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | D | Layouts can be manipulated in Unity. | https://arxiv.org/html/2408.09382v1 Section 3.1–3.3 |
| stereo | NR | No targeted evidence located in the inspected sections. | https://arxiv.org/html/2408.09382v1 Section 3.1–3.7; Section 4–5; Section 6.2 |
| temporal | NR | No targeted evidence located in the inspected sections. | https://arxiv.org/html/2408.09382v1 Section 3.1–3.7; Section 4–5; Section 6.2 |
| control | E | Perceived agency is measured across manual, scaffolded, and automatic creation in Study 2; it is not a geometric command-accuracy test. | https://arxiv.org/html/2408.09382v1 Sections 5.3-5.4 |
| interaction | E | User studies of pointing and object manipulation in VR. | https://arxiv.org/html/2408.09382v1 Section 4–5 |
| export | D | Existing furniture and generated layouts integrated in Unity. | https://arxiv.org/html/2408.09382v1 Section 3.7 |
| runtime | NR | No targeted evidence located in the inspected sections. | https://arxiv.org/html/2408.09382v1 Section 3.1–3.7; Section 4–5; Section 6.2 |
| xr | E | Two headset-based authoring user studies. | https://arxiv.org/html/2408.09382v1 Section 3.7; Section 4–5 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | D | Three-dimensional meshes generated and viewed in XR. | https://arxiv.org/html/2412.09008v1 Section III-C/D |
| stereo | NR | No targeted evidence located in the inspected sections. | https://arxiv.org/html/2412.09008v1 Section III–V |
| temporal | NR | No targeted evidence located in the inspected sections. | https://arxiv.org/html/2412.09008v1 Section III–V |
| control | D | Results shown for the same sketch with different spoken inputs. | https://arxiv.org/html/2412.09008v1 Section IV-B |
| interaction | D | Scaling, rotation, and translation through MRTK3. | https://arxiv.org/html/2412.09008v1 Section III-D |
| export | D | OBJ imported into Unity over HTTP. | https://arxiv.org/html/2412.09008v1 Section III-D |
| runtime | E | On a 4090: mean image generation time 3.83 s; mean mesh generation time 12.39 s. | https://arxiv.org/html/2412.09008v1 Section IV-A/B |
| xr | D | VR and MR use cases on Quest 3. | https://arxiv.org/html/2412.09008v1 Section IV-A; Section V |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | D | Unity scenes composed of 3DGS and meshes. | https://arxiv.org/html/2512.20129v1 Section 5.2; Section 5.5 |
| stereo | NR | No targeted evidence located in the inspected sections. | https://arxiv.org/html/2512.20129v1 Section 5–6; Section 8 |
| temporal | NR | No targeted evidence located in the inspected sections. | https://arxiv.org/html/2512.20129v1 Section 5–6; Section 8 |
| control | E | Control preferences compared between prompting and sculpting. | https://arxiv.org/html/2512.20129v1 Section 6.4.1 |
| interaction | E | Headset-based authoring tasks and interviews. | https://arxiv.org/html/2512.20129v1 Section 6 |
| export | D | Offline results returned to the Unity scene. | https://arxiv.org/html/2512.20129v1 Section 5.4–5.5 |
| runtime | C | Approximate stage times are reported for online previews, but a dedicated timing procedure or measured distribution was not located; full input-to-display latency remains NR. | https://arxiv.org/html/2512.20129v1 Sections 5.1.2-5.1.3 and 5.4.1 |
| xr | E | Study with 7 participants using Quest 3 with PC link. | https://arxiv.org/html/2512.20129v1 Section 6.1–6.4 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | D | GLB assets deployed in Unity scenes. | https://arxiv.org/html/2607.03731v1 Section 3.1.4 |
| stereo | NR | No targeted evidence located in the inspected sections. | https://arxiv.org/html/2607.03731v1 Section 3–5; Section 6.5–6.6 |
| temporal | NR | No targeted evidence located in the inspected sections. | https://arxiv.org/html/2607.03731v1 Section 3–5; Section 6.5–6.6 |
| control | D | The implemented dialogue supports prompt refinement and concept-image confirmation. Satisfaction, creativity-support ratings, and affective dominance do not specifically measure adherence to generative conditions or perceived authoring control; qualitative feedback identifies limits in scene-aware control. | https://arxiv.org/html/2607.03731v1 Sections 3.1.1-3.1.3, 4.4, 5.3, and 5.7 |
| interaction | E | Telemetry for grabbing, rotation, and scaling. | https://arxiv.org/html/2607.03731v1 Section 4.4; Section 5.4 |
| export | D | Runtime GLB loading with manipulation components added. | https://arxiv.org/html/2607.03731v1 Section 3.1.4 |
| runtime | E | A6000 endpoint configuration sweep; asynchronous queue. | https://arxiv.org/html/2607.03731v1 Section 3.2; Figure 2 |
| xr | E | Desktop web co-design followed by Quest 3 use; design and validation studies assess the specified workflow and generated scene outputs. | https://arxiv.org/html/2607.03731v1 Section 4.3–4.5 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Overlap-PSNR ratio evaluated across views. | https://arxiv.org/html/2307.01097v3 Sections5,5.3 |
| stereo | NR | No paired-eye or headset assessment located. | https://arxiv.org/html/2307.01097v3 Sections4–5 |
| temporal | NA | NA for temporal evolution in the core static image-set output. Spatial view interpolation and overlap correspondence do not constitute an ordered temporal process or a retained-state protocol. | https://arxiv.org/html/2307.01097v3 Section4 |
| control | E | CLIP text-image similarity evaluated. | https://arxiv.org/html/2307.01097v3 Section5/Table1 |
| interaction | NR | No input-driven runtime interaction located. | https://arxiv.org/html/2307.01097v3 Sections4–5 |
| export | D | Scene-mesh texture construction illustrated. | https://arxiv.org/html/2307.01097v3 Abstract/Figure1 |
| runtime | NR | Inference hardware stated; timed XR response not located. | https://arxiv.org/html/2307.01097v3 Section5 |
| xr | C | The broader-impact statement names generation of environments for virtual-reality experiences as an intended use. No headset implementation, device protocol, or user validation was located in the inspected material. | https://arxiv.org/html/2307.01097v3 Section 6 (Broader impact); Sections 1 and 4-5 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Spherical epipolar residuals are evaluated on retained rigid-scene correspondences after independently moving tracks are discarded. Pairs without recoverable two-view pose, including insufficient-parallax pairs, are excluded rather than assigned zero error. | https://arxiv.org/html/2609.38689v1 Sections 3.2 and 4.2; Section 4.3, Table 3; Section 5 limitations |
| stereo | E | Stereo PEGS changes from 3.121 to 2.640 mrad with 500 videos selected per stage. The score uses valid pose-recoverable pairs of retained rigid-scene correspondences after moving tracks are excluded; it is not a headset preference or comfort measurement. | https://arxiv.org/html/2609.38689v1 Sections 3.2 and 4.1–4.2; Section 4.3, Table 3 |
| temporal | E | Temporal PEGS improves, diagonal PEGS worsens, and temporal MEt3R worsens. PEGS evaluates retained rigid-scene correspondences after moving tracks and invalid low-parallax pairs are excluded; this does not measure consistency of the excluded object dynamics. | https://arxiv.org/html/2609.38689v1 Sections 3.2 and 4.2; Section 4.3, Table 3; Section 5 limitations |
| control | D | Prompt conditioning and stereo-baseline adjustment described. | https://arxiv.org/html/2609.38689v1 Sections3.1,3.4 |
| interaction | NR | Live action-driven manipulation not located. | https://arxiv.org/html/2609.38689v1 Sections3–4 |
| export | D | The playback conversion re-renders both omnidirectional-stereo eyes from the left view using depth whose scale is calibrated against the right view, then applies binocular super-resolution. This describes display-oriented processing, without a located device or user validation. | https://arxiv.org/html/2609.38689v1 Section 3.4, Equations 11–12; 4K Upsampling ordering statement |
| runtime | NR | End-to-end timing not located in inspected results. | https://arxiv.org/html/2609.38689v1 Section4 |
| xr | C | Headset suitability stated; device/user validation not located. | https://arxiv.org/html/2609.38689v1 Sections3.4,4.5 |
Spatial E concerns offline 14B experiments. Control, interaction and XR E refer to the deployed 5B study. These cells do not describe one jointly tested configuration.
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | The offline Wan2.2 14B experiment evaluates estimated 3D hand pose/shape and recovered camera trajectories. Table 1 reports PA-MPJPE 12.23 mm and PA-MPVPE 9.10 mm for hybrid conditioning; these are Procrustes-aligned hand-estimator proxies, not metric-scale scene geometry or multiview/stereo validation of the deployed 5B system. | https://arxiv.org/html/2602.18422v1 Section 4 Metrics; Tables 1-2; Appendix A.3 |
| stereo | NR | No paired-eye, IPD, or binocular-quality test was located in the inspected sections. Section 6 explicitly identifies stereo rendering among the system's limitations; headset use alone is not stereo validation. | https://arxiv.org/html/2602.18422v1 Sections 5-6; Appendix D |
| temporal | D | Short hand-object sequences are demonstrated. Appendix C deliberately uses short tasks to limit long-horizon drift, and Section 6 reports visual degradation after a few seconds. No intervention-departure-return test of retained object state was located; FVD is reported as distribution-level realism, not a persistent-state test. | https://arxiv.org/html/2602.18422v1 Section 4 Metrics; Sections 5-6; Appendices C-D |
| control | E | The 5B headset study compares tracked hand-plus-head conditioning with head-pose conditioning plus task text but no tracked hand control. Mean perceived control is 4.21 versus 1.74 on a seven-point scale. Offline 14B camera/hand-control metrics in Section 4 are a separate configuration. | https://arxiv.org/html/2602.18422v1 Section 4/Table 2; Section 5 User Study Design and Evaluating User Experience; Figure 7; Appendix C |
| interaction | E | Eleven participants perform three tasks, each within 8 seconds: press the green button, open a jar, and turn a steering wheel. Mean binary task accuracy is 71.2% with tracked hands versus 3.0% without them; generated session videos are blind-reviewed by a separate person. This tests short visual action outcomes, not persistent or physically verified object state. | https://arxiv.org/html/2602.18422v1 Section 5 User Study Design and Evaluating Task Efficiency; Figures 6-7; Appendix C |
| export | D | Unity on Quest 3 captures head/hand poses and receives generated video chunks from the server. This documents an engine/headset streaming pipeline; it does not demonstrate export or manipulation of editable 3D assets. | https://arxiv.org/html/2602.18422v1 Section 5 Integration with VR System |
| runtime | C | Section 5 reports 11 FPS, 1.4 s latency, and 0.002 s added conditioning overhead for the 5B system on one remote H100 with 12-frame chunks. A timing protocol, comparative timing experiment, or repeated-run distribution was not located. These author-reported figures are not an instrumented motion-to-photon result. | https://arxiv.org/html/2602.18422v1 Section 5 (Autoregressive Distillation; Integration with VR System); Appendix D |
| xr | E | Unity/Quest 3 deployment is evaluated with 11 participants on three short hand-object tasks and perceived control. Participants see streamed generated video; the observed task improvements do not establish binocular correctness, sustained comfort, or long-term state persistence. | https://arxiv.org/html/2602.18422v1 Section 5; Figures 6-7; Appendix C |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Simulator-reference frame fidelity; no metric 3D test. | https://arxiv.org/html/2603.06679v2 Section 4.3, Table 1 |
| stereo | NR | No calibrated paired-eye protocol located. | https://arxiv.org/html/2603.06679v2 Sections 3–6 |
| temporal | E | Reference fidelity over ranges 1–128 and 128–256; no arbitrary-edit retention test. | https://arxiv.org/html/2603.06679v2 Section 4.3, Table 1 |
| control | D | Authored map and action conditioning implemented. | https://arxiv.org/html/2603.06679v2 Sections 3.1–3.4, 4.2; Figure 3 |
| interaction | E | VLM opponent-presence agreement with simulator-frame judgments; one-map multiplayer evaluation. | https://arxiv.org/html/2603.06679v2 Sections 5.2–5.4, Table 2 |
| export | NR | No asset or engine export located. | https://arxiv.org/html/2603.06679v2 Sections 3–6 |
| runtime | C | Approximately 20 FPS reported, one A100 per player; not display latency. | https://arxiv.org/html/2603.06679v2 Section 5.3 |
| xr | NR | No AR/MR/VR device deployment located. | https://arxiv.org/html/2603.06679v2 Sections 3–6; project page |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Five scenes, 27 reference boxes, IoU ≥ 0.25; scene-graph recall/labels, not registration accuracy. | https://arxiv.org/html/2504.21360v2 Section 5.1, Table 2 |
| stereo | NR | No paired-eye evaluation located; handheld AR. | https://arxiv.org/html/2504.21360v2 Sections 4.3, 6–8 |
| temporal | D | Object-ID/transform updates and prior-asset reuse implemented; no longitudinal retention evaluation. | https://arxiv.org/html/2504.21360v2 Section 4.3 |
| control | E | Qualitative comparisons and participant reports examine perceived control, precision, and ownership during manual and AI-assisted authoring. This is a user-reported control target, not geometric constraint accuracy; general satisfaction or the creativity-support score alone is not used as control evidence. | https://arxiv.org/html/2504.21360v2 Sections 6.2-6.3 and 7.1-7.2 (authoring preferences and control reports) |
| interaction | E | Creation, placement, modification and removal studied in situ. | https://arxiv.org/html/2504.21360v2 Sections 6–7 |
| export | D | Generated textured meshes delivered into Unity mobile AR. | https://arxiv.org/html/2504.21360v2 Sections 4.2–4.3 |
| runtime | E | 50-trial component timing: total 33.92 ± 5.83 s; not display latency. | https://arxiv.org/html/2504.21360v2 Section 5, Table 1 |
| xr | E | 20 main participants, public-park handheld AR study; four pilots separate. | https://arxiv.org/html/2504.21360v2 Sections 6–7 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Three AI/manual object pairs undergo mesh-level geometry, topology, UV and organization inspection. These are technical properties of selected assets, not metric agreement with a historically verified 3D scene. | https://www.mdpi.com/2075-5309/16/16/3319 Sections 4.2.1-4.2.2; Tables 6-8 |
| stereo | NR | No paired-eye, interpupillary-distance, or binocular-quality protocol was located; PICO 4 presentation alone does not establish these properties. | https://www.mdpi.com/2075-5309/16/16/3319 Sections 4.1 and 4.3; Section 5.3 limitations |
| temporal | NR | The deployed scene workflow is examined for guided experience; no intervention-departure-return or retained-edit test was located. Static source content does not establish session-state persistence. | https://www.mdpi.com/2075-5309/16/16/3319 Sections 4.1 and 4.3; Section 5.3 |
| control | D | Structured prompts, manually selected historical references and human/expert correction form an implemented revision workflow. No standalone model diagnosis accuracy or condition-adherence test is claimed. | https://www.mdpi.com/2075-5309/16/16/3319 Sections 3.1 and 4.1.2; Figure 3; Section 5.3 |
| interaction | D | The Unity/PICO 4 workflow supports guided scene exploration. The paper explicitly lacks task-based usability testing; experience questionnaires do not establish manipulation accuracy or retained state. | https://www.mdpi.com/2075-5309/16/16/3319 Sections 4.1 and 4.3.1; Figure 7; Section 5.3 |
| export | D | Object outputs are imported into Blender for inspection, while scene outputs are deployed through Unity for VR presentation. This documents a processing/deployment route, not universal preservation of editable topology. | https://www.mdpi.com/2075-5309/16/16/3319 Sections 4.1 and 4.1.3; Section 4.2 |
| runtime | E | An exploratory three-pair workflow comparison records raw generation, manual modeling and use-dependent post-processing endpoints. Raw generation is approximately 3-4 min; close-up refinement 1-2 h; basic structural editability 4-5 h total post-processing, including close-up refinement. These are bounded workflow comparisons, not GPU inference or response-latency measurements. | https://www.mdpi.com/2075-5309/16/16/3319 Section 4.2.1, Table 6 (Build time); Section 4.2.2, Tables 7-8 and accompanying endpoint explanation |
| xr | E | Unity/PICO 4 presentation is evaluated using questionnaires and interviews with nine experts and 30 non-experts. Experts used a mixture of on-site VR and remote first-person video; their results are descriptive. The source describes guided non-expert VR experience but does not provide a participant-by-device log. | https://www.mdpi.com/2075-5309/16/16/3319 Sections 4.1, 4.3.1-4.3.3; Figure 7; Section 5.3 |
| Dimension | Code | Named property, configuration and limit | Source and locator |
|---|---|---|---|
| spatial | E | Straight-line distortion, seam quality and image-similarity measures assess the panoramic images. Expert ratings address perceived layout and visual consistency; these are image/perceptual proxies, not metric 3D reconstruction or translational-view validity. | https://www.mdpi.com/2075-5309/16/9/1679 Sections 3.5.2 and 4.3.4; Tables 3-7 |
| stereo | NR | No paired-eye generation, stereo baseline, interpupillary-distance manipulation or binocular-consistency protocol was located. Quest 3 viewing does not establish these properties. | https://www.mdpi.com/2075-5309/16/9/1679 Section 4.3.4; discussion and limitations |
| temporal | NA | NA for temporal evolution in the core static panoramic-stimulus output. The reported evaluation does not define an ordered-video or retained-modification task. | https://www.mdpi.com/2075-5309/16/9/1679 Section 3.1; Section 4.3.4; limitations |
| control | E | CLIP text-image alignment and expert prompt/greenery-alignment ratings evaluate specified conditioning targets. They do not establish precise three-dimensional placement or strict causal isolation of every environmental variable. | https://www.mdpi.com/2075-5309/16/9/1679 Section 3.5.2, Table 3; Section 4.3.4, Table 7 |
| interaction | NR | The inspected protocol establishes viewing of panoramic stimuli in a headset; a targeted input-driven manipulation, navigation-task or retained-state evaluation was not located. | https://www.mdpi.com/2075-5309/16/9/1679 Section 4.3.4; Tables 6-7 |
| export | D | Generated panoramic images are presented using Meta Quest 3 headsets. This documents delivery to a viewing configuration, not export of editable 3D geometry or a general engine-integration path. | https://www.mdpi.com/2075-5309/16/9/1679 Section 4.3.4 |
| runtime | NR | No generation, fixed-content rendering, or input-to-display timing protocol was located in the inspected sections. | https://www.mdpi.com/2075-5309/16/9/1679 Sections 3.2-3.5 and 4.3.4; discussion |
| xr | E | Seven senior experts view AI-generated and SketchUp/Enscape reference panoramas using Meta Quest 3 and rate presence and related perceptual qualities. The comparison concerns selected static stimuli; it does not establish free translation, binocular correctness or sustained comfort. | https://www.mdpi.com/2075-5309/16/9/1679 Section 4.3.4; Tables 6-7 |