# Tunnel Try-on: Excavating Spatial-temporal Tunnels for High-quality Virtual Try-on in Videos

**Authors:** Zhengze Xu, Mengting Chen, Zhao Wang, Linyu Xing, Zhonghua Zhai, Nong Sang, Jinsong Lan, Shuai Xiao, Changxin Gao  
**arXiv:** [2404.17571](https://arxiv.org/abs/2404.17571)  
**Date:** 2024-04-26  
**Summary page:** https://www.summarizepaper.com/en/paper/2404.17571v1/

## Key points

- Virtual try-on technology faces challenges in balancing clothing details and natural movements
- "Tunnel Try-on" framework introduced to address these challenges
- Tunnel Try-on creates spatial-temporal tunnels for close-up shots of clothing details
- Kalman filter used for smooth transitions and movements within focus tunnel
- Position embeddings injected into attention layers for continuity in videos
- Environment encoder extracts contextual information for immersive experience
- Tunnel Try-on maintains fidelity to clothing details and realistic movements in videos
- Represents a significant advancement in virtual try-on technology for commercial applications

## Summary

In the realm of virtual try-on technology, video try-on presents a unique set of challenges that have yet to be effectively addressed in previous research efforts. The primary hurdle lies in finding a balance between preserving intricate clothing details and ensuring natural and coherent movements and interactions with the garments. To overcome these obstacles, a groundbreaking diffusion-based framework known as "Tunnel Try-on" has been introduced. This innovative concept involves creating spatial-temporal tunnels within the input video footage to focus on specific regions around the clothing for close-up shots. By honing in on these areas, Tunnel Try-on can meticulously preserve even the finest details of the clothing items being virtually tried on. To further enhance realism and fluidity of motion in generated videos, advanced techniques are employed. One such technique is using the Kalman filter to construct smooth crops within the focus tunnel, ensuring seamless transitions and movements. Additionally, position embeddings from the tunnel are strategically injected into attention layers to maintain continuity throughout the videos. Furthermore, an environment encoder is utilized to extract contextual information from outside the tunnels, providing supplementary cues for a more immersive virtual try-on experience. The culmination of these cutting-edge techniques results in Tunnel Try-on's ability to maintain exceptional fidelity to clothing details while synthesizing stable and smooth videos that accurately reflect realistic movements and interactions. This significant advancement positions Tunnel Try-on as a pioneering step towards achieving commercial-level applications of virtual try-on technology in videos. With its unparalleled capabilities and innovative approach, Tunnel Try-on sets a new standard for high-quality virtual try-on experiences that push boundaries and redefine possibilities in digital fashion engagement.