Moving Beyond More Views:
Redundancy-Aware Ego–Exo Fusion for Proficiency Estimation

1 University of Surrey   2 University of Wollongong
European Conference on Computer Vision (ECCV), 2026

Are more camera views always better?

No. In Ego–Exo proficiency estimation, redundant or noisy exocentric views can dilute useful motion cues and increase overfitting.

We learn both which views to look at and how to fuse them. AdaMVS selects informative views, while VIB-GB regularises the fused representation to suppress view-specific overfitting.

Multi-exo-view fusion challenge and AdaMVS view selection

Conventional fusion treats all viewpoints as useful. Our analysis shows that some exocentric views contribute little information, and adding views can even reduce accuracy. AdaMVS instead learns to prioritise informative views before fusion.

Abstract

Ego–Exo proficiency estimation combines fine-grained motion cues from egocentric video with spatial context from multiple exocentric cameras. However, simply adding more exocentric views can degrade performance because redundant or noisy perspectives dilute discriminative information and enlarge the feature space, encouraging view-specific overfitting. We address these problems with two complementary components. AdaMVS adaptively identifies and fuses the most informative view tokens under weak supervision, while VIB-GB combines Variational Information Bottleneck regularisation with Gradient Blending to compress redundant signals and balance learning across ego and exo branches. Experiments on EgoExo-4D and EgoExo-Fitness show that the framework learns both which view to look at and how to fuse it, achieving state-of-the-art proficiency estimation with improved robustness and efficiency.

Two problems, two complementary solutions

1. Multiview redundancy

Different exocentric cameras often contain duplicated, occluded, blurred, or weakly informative content.

AdaMVS

Scores view tokens, selects the most informative Top-K exocentric views, and performs adaptive token-level fusion without explicit view-level labels.

2. Fusion overfitting

More views enlarge the representation and can cause one branch to dominate, memorising view-specific patterns rather than generalisable cues.

VIBGradient Blending

Compresses non-essential latent information and dynamically balances ego/exo learning using overfitting-aware gradient weighting.

Architecture

AdaMVS and VIB-GB architecture

The egocentric stream captures fine-grained motion cues, while Exo-AdaMVS scores and selects informative exocentric views. Ego–Exo AdaMVS then fuses the streams, and VIB-GB regularises training to reduce overfitting before ordinal proficiency estimation.

Key results

53.0%
Best Ego+Exo accuracy on EgoExo-4D (K400)
+5.5
percentage-point gain over SkillFormer on EgoExo-4D
48.8%
Ego+Exo accuracy on EgoExo-Fitness (K600)

Across pretrained backbones, the proposed framework consistently improves over conventional Ego–Exo fusion. On EgoExo-4D, AdaMVS reaches 53.0% with K400 features, compared with 47.5% for SkillFormer. The ablation study also shows the complementarity of the two ideas: AdaMVS improves the ego–exo baseline from 45.5% to 48.0%, while the complete AdaMVS + CORN + VIB-GB configuration reaches 51.0% in the controlled ablation setting.

Efficiency

0.26
GFLOPs — AdaMVS-Small
2.19M
parameters — AdaMVS-Small

AdaMVS-Small achieves 50.1% Ego+Exo accuracy with an over 11× reduction in computation and model size relative to AdaMVS-Large, retaining strong performance while substantially reducing complexity.

What does AdaMVS actually select?

Qualitative AdaMVS view selection results

In successful cases, high-scoring views provide clearer and more discriminative cues, while low-scoring views are often affected by occlusion, blur, or redundant content. The model selects the Top-2 exocentric views and prunes the rest. Failure cases occur when viewpoints contain highly overlapping information and receive similar scores.

Datasets

EgoExo-4D

1,087 hours of multiview video across nine activity categories, captured using one egocentric and four exocentric views, with proficiency labels across four discrete levels.

EgoExo-Fitness

1,276 cross-view videos (about 32 hours), segmented into 6,131 single actions and annotated with interpretable action judgements and five-level quality scores.

Want to explore the details?

Read the ECCV paper or explore the implementation.

Frequently Asked Questions

Why can adding more exocentric views hurt?

Extra cameras may add duplicated, noisy, blurred, or occluded observations. Static fusion gives these weak views too much influence and also increases the capacity available for view-specific overfitting.

Does AdaMVS need labels saying which camera is best?

No. View importance is learned weakly from the proficiency-estimation task loss, without explicit view-level supervision.

What is the role of VIB-GB?

VIB compresses redundant latent information, while Gradient Blending reweights ego and exo learning according to branch-level overfitting. Together they improve generalisation after view selection.

Can the method work with a different number of views?

The selector is designed to be view-number agnostic through shared queries and permutation-invariant Transformer processing, allowing it to adapt to varying numbers of exocentric views.

BibTeX

@inproceedings{Dong2026AdaMVS,
  author    = {Xu Dong and Wanqing Li and Anthony Adeyemi-Ejeye and Andrew Gilbert},
  title     = {Moving Beyond More Views: Redundancy-Aware Ego--Exo Fusion for Proficiency Estimation},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}