AVA-VLM: Adaptive Visual Attention-Vision Language Model for In-the-Wild Construction Site Monitoring

2026-07-07

AI Summary

This work addresses key limitations of current vision-language models in real-world construction site monitoring—namely limited effective range, poor reliability under low-resolution conditions, and inefficient inference. To overcome these challenges, the authors propose a human cognition-inspired coarse-to-fine reasoning framework that first performs an initial assessment using low-resolution global imagery and adaptively requests high-resolution crops of specific regions only when necessary for fine-grained analysis. By integrating an adaptive visual attention mechanism with a region-aware chain-of-thought training strategy, the model learns to dynamically determine whether to inspect further, where to crop, and how to fuse multi-scale evidence. This approach significantly improves monitoring accuracy in long-range, low-resolution scenarios while substantially reducing visual token consumption, thereby enabling both efficient and reliable reasoning.

Abstract

Vision-Language Models (VLMs) are promising for construction-site monitoring, and recent construction-tailored VLMs have primarily adapted pretrained VLMs through direct QA-style fine-tuning from a single global image. We argue that this direct paradigm remains limited for in-the-wild deployment in terms of operational range, reliability under reduced-resolution inputs, and inference efficiency. To address these challenges, we propose AVA-VLM, an Adaptive Visual Attention-Vision Language Model that follows a human-inspired coarse-to-fine reasoning strategy. AVA-VLM first reasons over a low-resolution global image and selectively requests a high-resolution local crop only when detailed inspection is needed, similar to how a human inspector zooms in on hard-to-see yet important areas. We further introduce a region-aware Chain-of-Thought dataset that teaches the model when to inspect, where to crop, and how to use local evidence. Experiments show that AVA-VLM improves reliability under long-distance and reduced-resolution conditions while substantially reducing visual-token usage.

Problem

• construction-site monitoring
• Vision-Language Models
• in-the-wild deployment
• reduced-resolution inputs
• inference efficiency

Innovation

• Adaptive Visual Attention
• Coarse-to-Fine Reasoning
• Vision-Language Model
• Region-aware Chain-of-Thought
• In-the-Wild Monitoring