As long-video understanding gains increasing attention, augmented reality (AR) devices are emerging as a compelling deployment platform. However, egocentric videos exhibit significant temporal redundancy, imposing substantial memory and computational burdens on resource-constrained AR hardware. To address this challenge, we introduce UTOPI, a plug-and-play pre-compression module designed specifically for egocentric long videos on AR platforms. First, we leverage user motion signals to estimate inter-frame overlap and propose a viewpoint-shift-invariant pruning algorithm. Second, we exploit eye-tracking sensors to identify user fixations, enabling the separation of attended foreground content from background and its preferential retention via a fine-tuned neural network-based adjustment mechanism. Across multiple benchmarks and model backbones, UTOPI achieves up to 95% compression while preserving performance. Code and project page are available at https://sai-lab-nyu.github.io/utopi/.
Figures & tables
Figure 1: (a) UTOPI compresses temporal redundancy from viewpoint changes and spatial redundancy in unattended regions. (b) An example of how user attention varies with gaze.
Figure 2: (a) Meta Aria glasses equipped with multiple cameras. (b) Meta Quest Pro headset, which functions as both an AR and VR device. (c) Distribution of image differences.
Figure 3: The overall workflow of UTOPI.
Figure 4
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 6: Additional question-option examples in EgoEye and their corresponding selected question subjects.
Figure 6Figure 7
Figure 11: The visualization of re-projection. The odd rows show the original sampled frames, while the even rows show the previous sampled frames re-projected onto the current frames. To better illustrate the re-mapping quality, we present pixel-level re-projection here. As can be seen, the frames obtained by mapping from the previous frame closely match the original frames. The differences between the upper and lower frames mainly come from two factors: (1) the black regions indicate locations that are not visible in the previous frame; (2) dynamic contents, such as Frame 27 in the second video and the frame labeled “from 18”.