What if, whereas watching The Dark Knight, you were not simply observing the Joker on display screen, however truly seeing Gotham City by way of his eyes? The video know-how that permits viewers to expertise the world by way of a personality’s perspective, fairly than as a mere observer, is turning into a actuality. Researchers at our college have developed a brand new AI mannequin that generates first-person viewpoint movies from commonplace footage.
KAIST introduced on February twenty third that Professor Jaegul Choo’s analysis staff on the Kim Jaechul Graduate School of AI has developed ‘EgoX,’ an AI mannequin that makes use of observer-perspective (exocentric) video to exactly generate the scenes that an individual within the video would truly be seeing.
With the fast development of Augmented Reality (AR), Virtual Reality (VR), and AI robotics, the significance of “egocentric video”—which captures scenes as one immediately sees them—is rising. However, acquiring high-quality first-person footage beforehand required customers to put on costly motion cameras or sensible glasses. Furthermore, there have been vital technical limitations in naturally changing present commonplace (third-person or exocentric) video right into a first-person perspective.
A key characteristic of this know-how is that it goes past merely rotating the display screen; it comprehensively understands the individual’s place, posture, and the 3D construction of the encompassing house to reconstruct the first-person viewpoint.
Existing applied sciences usually solely transformed nonetheless photos or required footage from 4 or extra cameras. Additionally, they steadily suffered from awkward visible artifacts in movies with complicated lighting or fast motion.
In distinction, EgoX can generate high-quality first-person video from only a single third-person video supply. Specifically, the analysis staff succeeded in realistically implementing pure shifts in imaginative and prescient—akin to when an individual turns their head—by exactly modeling the correlation between head motion and the precise subject of view.
This know-how demonstrated steady efficiency throughout varied each day situations, together with cooking, exercising, and dealing, with out being restricted to particular environments. It is being evaluated as a breakthrough that opens new potentialities for securing high-quality first-person knowledge from present video archives with out the necessity for wearable gadgets.
EgoX is predicted to have a big affect throughout varied industries. In the fields of AR, VR, and the Metaverse, it could maximize person expertise by remodeling commonplace movies into immersive content material that makes customers really feel as if they’re experiencing the scene firsthand.
Furthermore, it’s projected to contribute to the fields of robotics and AI coaching by serving as core knowledge for “Imitation Learning,” the place robots be taught by watching human actions. New kinds of video companies, akin to switching sports activities broadcasts or vlogs to the attitude of the athlete or the protagonist, are additionally anticipated.
Distinguished Professor Jaegul Choo acknowledged, “This research is significant in that AI has moved beyond simple video conversion to learning and reconstructing human ‘vision’ and ‘spatial understanding.’ We expect an environment to open up where anyone can create and experience immersive content using only previously recorded videos.” He added, “KAIST will continue to secure global competitiveness in the field of generative AI-based video technology.”
This analysis was led by first authors Taewoong Kang, Kinam Kim, and Dohyeon Kim . The paper was pre-released on arXiv on December 9, 2025, garnering vital consideration from AI business giants like NVIDIA and Meta, in addition to academia. It is scheduled for official presentation on the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), a global tutorial convention to be held in Colorado, USA, on June 3, 2026.
Paper Title: EgoX: Egocentric Video Generation from a Single Exocentric Video Paper Link: https://keh0t0.github.io/EgoX/
Meanwhile, this analysis was supported by the Ministry of Science and ICT by way of the National Research Foundation of Korea’s particular person primary analysis venture, “Research on User-Centered Content Generation and Editing Technology through Generative AI,” and the Supercomputer No. 5 High-Performance Computing-based R&D Innovation Support venture, “Research on Video Filming Viewpoint Conversion Based on Diffusion Models.”