~*~ Multi-Space NeRF / Research notes & token protocol ~*~

# MULTI-SPACE NEURAL RADIANCE FIELDS

Parallel feature fields for reflective and refractive scenes. A study of neural scene representations beyond a single radiance field.

Conventional NeRF formulations struggle with mirror-like surfaces and complex light paths. Multi-Space NeRF introduces parallel feature fields in virtual sub-spaces, with modest computational overhead and compatibility with established NeRF architectures [1].

X Community   /   X   /   Max S website

~*~

## RESEARCH SCOPE

+--------------------------+--------------------------------+
| Research question        | Working frame                  |
+--------------------------+--------------------------------+
| Reflective surfaces      | Parallel virtual sub-spaces    |
| Novel-view synthesis     | Multi-view scene consistency   |
| Baseline comparison      | NeRF / Mip-NeRF / Mip-NeRF 360 |
| Evaluation setting       | 360-degree scene reconstruction|
+--------------------------+--------------------------------+

## HOW IT WORKS

Neural radiance fields model a scene as a continuous function of spatial position and viewing direction [2]. A single field can produce inconsistent geometry around mirror-like objects, where the apparent scene changes with the reflection path.

The multi-space formulation represents the scene through a collection of feature fields over virtual sub-spaces [1]. The objective is to account for reflected and refracted observations without requiring manually labelled reflective regions.

The following is a conceptual summary, not executable code or a reproduction of the original implementation:

input  = calibrated multi-view images
space  = { F_0, F_1, ... F_k }  // parallel feature fields
model  = multi-space radiance representation
output = rendered novel viewpoints

// See Yin et al. [1] for the full formulation.

## EVALUATION CONTEXT

The original CVPR 2023 MS-NeRF paper reports evaluation across 25 synthetic scenes and 7 real captured scenes, including reflection and refraction under 360-degree viewpoints [1]. These figures refer to published research, not an evaluation conducted by this project.

+--------------------------+--------------------------+
| Published evaluation     | Reported setting         |
+--------------------------+--------------------------+
| Synthetic scenes         | 25                       |
| Real captured scenes     | 7                        |
| Camera viewpoints        | 360 degrees              |
| Model compatibility      | Three NeRF baselines     |
+--------------------------+--------------------------+

Version note: the figures above refer to CVPR 2023 (arXiv v1). The updated TPAMI 2025 version expands the synthetic dataset to 33 scenes.

Research dataset — Hugging Face →

## TOKEN PROTOCOL

$MSNERF is a proposed coordination token for an independent, research-oriented initiative. Its intended role is to organize support for open datasets, reproducible benchmarks, and shared compute, rather than to create ownership of the underlying academic work.

The token layer and the cited method are distinct. The academic papers do not describe, authorize, or endorse this issuance proposal.

## PROPOSED UTILITY

• Research grants tied to specified milestones and public results
• Community review of datasets, benchmarks, and technical proposals
• Transparent accounting for compute and evaluation expenditures
• Public records of protocol decisions and treasury movements

## ISSUANCE STATUS

The issuance design remains under specification. No contract, supply figure, distribution ratio, or launch date is asserted in this concept preview.

+--------------------------+--------------------------------+
| Field                    | Current proposal               |
+--------------------------+--------------------------------+
| Working symbol           | $MSNERF                        |
| Network                  | To be specified                |
| Total supply             | Not announced                  |
| Contract address         | Not deployed                   |
| Distribution schedule    | Draft pending                  |
| Treasury reporting       | Public accounting proposed     |
+--------------------------+--------------------------------+

## GOVERNANCE PRINCIPLES

Any governance mechanism should be evaluated against clear operational questions: what resources can be allocated, which decisions are reviewable, and what constitutes a verifiable research deliverable. Token-based voting, if implemented, would require separately published parameters and safeguards.

Participation in the project should not be confused with authorship of the cited work. Research claims remain attributable to their original publications.

## REFERENCES

## ABOUT

This independent project studies Multi-Space NeRF as a technical reference point for open 3D representation infrastructure. The proposed token mechanism is an organizational experiment, not a result of the cited research program.

References are provided for scholarly attribution. No affiliation, partnership, or endorsement by the authors, their institutions, or the operators of the referenced websites is implied.

~*~