The Spatial Scale Continuum in Visual Perception

14 concepts5 papers

For decades, cognitive science and computer vision operated under a clean, binary assumption: the visual world consists of isolated objects we manipulate and large-scale scenes we navigate. If you look at a coffee mug, your brain processes it as an object; if you look at an entire kitchen, it processes it as a scene.

However, this dichotomy leaves a massive gap in our understanding of daily life. What happens when you look at a kitchen counter or a workbench? These environments contain multiple objects, yet we do not navigate through them. Instead, they serve as intermediate spaces optimized for manual interaction.

Recent insights across visual neuroscience and artificial intelligence suggest that the visual world is not a collection of discrete categories, but rather a continuous spatial-scale manifold. By shifting from a binary framework to a continuous scale continuum, researchers are uncovering how both human brains and machine vision models achieve fluid, fine-grained perception.

Beyond the Object-Scene Dichotomy

The realization that intermediate spaces form their own distinct visual domain began by challenging traditional psychological frameworks. In Perceptual dissociations among views of objects, scenes, and reachable spaces, Emilie Josephs demonstrated that these "reachspaces"—near-scale environments within arm's reach—are treated by both human perception and deep neural networks as a distinct category with its own unique visual signature. Rather than clustering tightly with either objects or scenes, reachspaces occupy a stable, intermediate position along a continuous scale spectrum.

To determine how the brain handles this intermediate scale, Emilie Josephs conducted fMRI studies detailed in Large-scale dissociations between views of objects, scenes, and reachable-scale environments in visual cortex. The study identified a specialized network of regions in the visual cortex specifically tuned to reachable-scale environments. These regions exhibit activation patterns distinct from the well-known pathways dedicated to isolated objects or expansive landscapes.

To provide the empirical foundation for studying this scale, Emilie Josephs et al. introduced the Reachspace Database in The world within reach: An image database of reach-relevant environments. By curating over 10,000 first-person images of environments within arm's reach, the authors established that our immediate interactive surroundings require dedicated computational and cognitive models separate from broad navigation or single-object recognition.

How the Brain Encodes the Continuum

If the visual cortex recognizes distinct scales, how does it transition smoothly between them without jarring perceptual jumps? Real-world vision is fluid; as we walk toward a desk, an expansive scene shrinks into a reachspace, which then focuses into a single object.

The mechanism behind this fluid transition was uncovered by Jeongho Park et al. in Ramp-shaped neural tuning supports graded population-level representation of the object-to-scene continuum. Instead of using rigid categorical boundaries, large areas of the visual cortex employ "ramp-shaped" neural tuning.

Categorical Tuning (Rigid Boundaries)
Scale:  [ Object ] -> [ Reachspace ] -> [ Scene ]
Neural: [  On/Off ]    [   On/Off   ]    [ On/Off ]

Ramp-Shaped Tuning (Continuous Gradient)
Scale:  Small --------------------------> Large
Neural: Neuron A:  \___________________________ (Decreasing activity)
        Neuron B:  ___________________________/ (Increasing activity)

In a ramp-shaped tuning model, individual neurons do not act as binary switches for specific categories. Instead, their activity levels increase or decrease monotonically as the spatial scale grows or shrinks. When aggregated across a large population of neurons, these individual responses form a smooth, linear population code. This continuous gradient allows the visual cortex to represent the entire spatial continuum seamlessly, ensuring that our perception shifts naturally as our physical distance from a target changes.

Translating the Continuum to Artificial Intelligence

This insight—that spatial scale is a continuous manifold rather than a set of discrete categories—is now solving critical bottlenecks in computer vision, particularly for Multimodal Large Language Models (MLLMs).

When processing ultra-high-resolution images, MLLMs frequently struggle with fine-grained perception. If an image contains a tiny object occupying a fraction of a percent of the frame, standard models fail to detect it. Traditional computer vision fixes this by breaking the image into a discrete grid of patches and zooming in. However, rigid grids introduce a major flaw: they frequently bisect target objects right along patch boundaries, destroying the integrity of visual features.

To overcome this, Geng Li et al. developed Bayesian Visual Search (BVS) in BVS: Bayesian Visual Search with Multimodal Large Language Model for Fine-grained Perception. Instead of relying on a rigid, discrete grid, the authors modeled the search space as a continuous spatial-scale manifold.

By treating the image scale as a continuous manifold, the model can dynamically adjust its focus and crop boundaries with infinite flexibility. Objects are never arbitrarily split across patch borders because the cropping mechanism operates smoothly across the scale continuum. This grid-free, boundary-aware visual sampling preserves fine-grained features during multi-stage zooming, enabling robust perception of tiny objects in massive visual fields by mirroring the continuous nature of biological vision.

Key Foundational Work

Large-scale dissociations between views of objects, scenes, and reachable-scale environments in visual cortex (2020) — contributed 5 concept elements to this direction.
Perceptual dissociations among views of objects, scenes, and reachable spaces. (2019) — contributed 4 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_spatial_scale_continuum_a148d68a,
  title = {spatial scale continuum},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/spatial-scale-continuum/rdr_a148d68a3d9f4d6bbf8be826c20bb67e},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_spatial_scale_continuum_a148d68a_v1,
  title = {spatial scale continuum},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/spatial-scale-continuum/rdr_a148d68a3d9f4d6bbf8be826c20bb67e/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}

See something wrong? Report a mistake