Title
Proposal for Proteus Temporal Natural: A Temporally‑Adaptive Refinement Model for High‑Stability Video Upscaling
Proposed Model Name
Proteus Temporal Natural (PTN)
or
Proteus Natural Temporal Adaptive (PNTA)
Both names reflect the model’s purpose:
Natural refinement + temporal coherence + adaptive processing.
1. Introduction
Current Proteus models operate primarily on a spatial basis.
While Proteus Natural already performs adaptive enhancement based on local sharpness, it does not incorporate temporal coherence analysis.
This proposal introduces a temporal‑adaptive refinement layer designed to stabilize:
-
motion blur
-
out‑of‑focus regions
-
background gradients
-
low‑frequency textures
-
secondary faces
-
soft edges
The goal is to eliminate temporal oscillations and micro‑vibrations while preserving cinematic softness.
2. Problem Statement
Even with Proteus Natural, certain artifacts persist:
-
flickering in soft backgrounds
-
instability in motion‑blurred regions
-
oscillation of fine textures
-
inconsistent enhancement of out‑of‑focus faces
-
temporal “breathing” of gradients
These issues arise because the model lacks temporal awareness.
3. Proposed Solution — Temporal Focal Coherence Mask (TFCM)
The core innovation is a temporal focal coherence mask, computed per frame and propagated across time.
Mask Values
-
0.0 → fully protected region
(motion blur, out‑of‑focus, soft backgrounds) -
1.0 → fully enhanced region
(stable sharp detail) -
0.0–1.0 → adaptive enhancement
(regions with mixed temporal behavior)
4. How the Temporal Mask Works
Step 1 — Spatial Sharpness Map
Same principle as the existing Proteus Natural logic.
Detects sharp vs. soft regions.
Step 2 — Temporal Stability Map
Analyzes each region across N frames to determine:
-
consistency of sharpness
-
persistence of blur
-
motion vector coherence
-
gradient stability
-
texture continuity
Step 3 — Temporal Focal Coherence Mask (TFCM)
Combines spatial and temporal data:
TFCM=f(Sharpness.spatial,Stability.temporal)
Step 4 — Adaptive Inference
Proteus enhancement strength is modulated by TFCM:
-
stable sharp regions → full enhancement
-
stable soft regions → no enhancement
-
unstable regions → partial enhancement
5. Expected Benefits
1. Perfectly stable motion blur
No more flickering or “breathing”.
2. Cinematic depth of field preservation
Out‑of‑focus regions remain consistent across frames.
3. Background stability
Soft gradients no longer oscillate.
4. Improved facial hierarchy
Secondary faces remain soft and natural.
5. No temporal over‑sharpening
Eliminates the “crispy flicker” effect.
6. Ideal for SD, analog, and interlaced sources
Where temporal instability is most visible.
6. Implementation Notes
-
The temporal mask can be computed using optical flow or motion vectors.
-
The mask should be smoothed temporally to avoid abrupt transitions.
-
The model should expose a user parameter:
Temporal Adaptivity Strength (0–100) -
The system must remain compatible with 10‑bit pipelines.
7. Conclusion
Proteus Temporal Natural (PTN) would represent a major evolution of the Proteus family, combining:
-
spatial adaptivity
-
temporal coherence
-
optical realism
-
cinematic rendering
This model would dramatically improve restoration quality, especially for SD and analog sources.
Cheers, Vincent.
