O-MaMa: Learning Object Mask Matching between Egocentric and Exocentric Views

TL;DR

O-MaMa introduces a novel mask matching approach for cross-view object segmentation, leveraging multi-perspective feature fusion and contrastive learning to improve accuracy in egocentric and exocentric views.

Contribution

It presents a new framework combining semantic feature pooling, cross-attention, contrastive loss, and hard negative mining for cross-view object segmentation.

Findings

Achieves state-of-the-art results on Ego-Exo4D benchmark.

Significant relative gains in IoU over baselines.

Efficient with only 1% of training parameters.

Abstract

Understanding the world from multiple perspectives is essential for intelligent systems operating together, where segmenting common objects across different views remains an open problem. We introduce a new approach that re-defines cross-image segmentation by treating it as a mask matching task. Our method consists of: (1) A Mask-Context Encoder that pools dense DINOv2 semantic features to obtain discriminative object-level representations from FastSAM mask candidates, (2) an Ego$\leftrightarrow$Exo Cross-Attention that fuses multi-perspective observations, (3) a Mask Matching contrastive loss that aligns cross-view features in a shared latent space, and (4) a Hard Negative Adjacent Mining strategy to encourage the model to better differentiate between nearby objects. O-MaMa achieves the state of the art in the Ego-Exo4D Correspondences benchmark, obtaining relative gains of +22% and +76% in the Ego2Exo and Exo2Ego IoU against the official challenge baselines, and a +13% and +6% compared with the SOTA with 1% of the training parameters.