The year/Independent research

Paper 2512.08269

EgoX: Egocentric Video Generation from a Single Exocentric Video

Published
Dec 2025
Research lab
Independent
Citations
6
GitHub
741 stars

01 In brief

Summary

EgoX is a novel framework that generates egocentric (first-person) videos from a single exocentric (third-person) video input.

It leverages a pretrained video diffusion model (Wan 2.1 14B) with lightweight LoRA adaptation, avoiding the need for additional inputs like multiple views or initial frames.

The method lifts the exocentric video into a 3D point cloud, renders an egocentric prior video, and combines it with the exocentric latent via width-wise and channel-wise concatenation.

A geometry-guided self-attention mechanism uses 3D direction vectors to focus on spatially relevant regions, suppressing unrelated content.

Experiments on Ego-Exo4D show EgoX outperforms baselines (Exo2Ego-V, TrajectoryCrafter, Wan Fun Control, Wan VACE) in image, object, and video metrics, with ablations confirming each component's contribution.

The framework generalizes to unseen and in-the-wild videos, though it requires an egocentric camera pose as input, suggesting future work on automatic head-pose estimation.

The paper demonstrates significant improvements in geometric consistency and visual fidelity for exo-to-ego video generation.

02 From the paper

Abstract

Egocentric perception enables humans to experience and understand the world directly from their own point of view. Translating exocentric (third-person) videos into egocentric (first-person) videos opens up new possibilities for immersive understanding but remains highly challenging due to extreme camera pose variations and minimal view overlap. This task requires faithfully preserving visible content while synthesizing unseen regions in a geometrically consistent manner. To achieve this, we present EgoX, a novel framework for generating egocentric videos from a single exocentric input. EgoX leverages the pretrained spatio temporal knowledge of large-scale video diffusion models through lightweight LoRA adaptation and introduces a unified conditioning strategy that combines exocentric and egocentric priors via width and channel wise concatenation. Additionally, a geometry-guided self-attention mechanism selectively attends to spatially relevant regions, ensuring geometric coherence and high visual fidelity. Our approach achieves coherent and realistic egocentric video generation while demonstrating strong scalability and robustness across unseen and in-the-wild videos.