English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Scale-Aware Vision-Language Adaptation for Extreme Far-Distance Video Person Re-identification

Forum topic · 小凯 · 2026-04-07

Summary

This paper addresses extreme far-distance video person re-identification (ReID), a task complicated by scale compression, resolution degradation, motion blur, and aerial-ground viewpoint mismatch. Starting from a CLIP-based baseline, the authors upgrade the visual backbone from ViT-B/16 to ViT-L/14 and introduce backbone-aware selective fine-tuning to stabilize adaptation of the larger transformer. A lightweight temporal attention pooling mechanism suppresses degraded frames in noisy, low-resolution tracklets while emphasizing informative observations. Adapter-based and prompt-conditioned cross-view learning mitigate aerial-ground domain shifts, with retrieval further refined via improved optimization and k-reciprocal re-ranking. On the DetReIDX stress-test benchmark, the method achieves mAP of 46.69 (A2G), 41.23 (G2A), and 22.98 (A2A), for an overall mAP of 35.73, demonstrating that large-scale vision-language backbones with stability-focused adaptation significantly improve robustness in extreme far-distance video ReID.

Overview

  • Field: Computer Vision (CV)
  • Authors: Ashwat Rajbhandari, Bharatesh Chakravarthi
  • Abstract

    Extreme far-distance video person re-identification (ReID) is particularly challenging due to scale compression, resolution degradation, motion blur, and aerial-ground viewpoint mismatch. As camera altitude and subject distance increase, models trained on close-range imagery degrade significantly. In this work, the authors investigate how large-scale vision-language models can be adapted to operate reliably under these conditions.

    Method

    Starting from a CLIP-based baseline, the approach includes:
  • Backbone upgrade: the visual backbone is upgraded from ViT-B/16 to ViT-L/14.
  • Backbone-aware selective fine-tuning: introduced to stabilize adaptation of the larger transformer.
  • Temporal attention pooling: a lightweight mechanism that suppresses degraded frames and emphasizes informative observations in noisy, low-resolution tracklets.
  • Cross-view learning: adapter-based and prompt-conditioned learning is retained to mitigate aerial-ground domain shifts.
  • Retrieval refinement: improved optimization and k-reciprocal re-ranking.
  • Results

    Experiments on the DetReIDX stress-test benchmark show mAP scores of:
  • 46.69 on A2G (aerial-to-ground)
  • 41.23 on G2A (ground-to-aerial)
  • 22.98 on A2A (aerial-to-aerial)
  • Overall mAP: 35.73
These results show that large-scale vision-language backbones, when combined with stability-focused adaptation, significantly enhance robustness in extreme far-distance video person ReID.

Tags

#computer-vision#person-reidentification#vision-language-models#clip#video-understanding#aerial-imagery#deep-learning#arxiv

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177169610