Slow Perception: Let's Perceive Geometric Figures Step-by-step
Haoran Wei, Youyang Yin, Yumeng Li, Jia Wang, Liang Zhao, Jianjian Sun, Zheng Ge, Xiangyu Zhang
2024-12-31
Summary
This paper talks about a new approach called 'slow perception' (SP) that helps AI models understand geometric figures step-by-step, similar to how humans do.
What's the problem?
Current large vision language models (LVLMs) struggle with visual reasoning tasks, especially when it comes to understanding complex geometric shapes. They often fail to accurately replicate these figures and do not grasp the underlying logic and relationships between different shapes, which limits their effectiveness in solving geometric problems.
What's the solution?
To tackle this issue, the authors introduce the concept of slow perception, which breaks down the process of understanding geometric figures into manageable steps. The method has two main stages: first, it decomposes complex shapes into simpler parts to make them easier to analyze; second, it uses a 'perceptual ruler' to trace lines carefully without skipping over important details. This gradual approach allows the model to build a better understanding of the shapes as it processes them.
Why it matters?
This research is important because it provides a more human-like way for AI to perceive and understand geometry. By teaching models to analyze shapes step-by-step, we can improve their ability to solve visual reasoning tasks, making them more useful in fields like education, robotics, and computer graphics.
Abstract
Recently, "visual o1" began to enter people's vision, with expectations that this slow-thinking design can solve visual reasoning tasks, especially geometric math problems. However, the reality is that current LVLMs (Large Vision Language Models) can hardly even accurately copy a geometric figure, let alone truly understand the complex inherent logic and spatial relationships within geometric shapes. We believe accurate copying (strong perception) is the first step to visual o1. Accordingly, we introduce the concept of "slow perception" (SP), which guides the model to gradually perceive basic point-line combinations, as our humans, reconstruct complex geometric structures progressively. There are two-fold stages in SP: a) perception decomposition. Perception is not instantaneous. In this stage, complex geometric figures are broken down into basic simple units to unify geometry representation. b) perception flow, which acknowledges that accurately tracing a line is not an easy task. This stage aims to avoid "long visual jumps" in regressing line segments by using a proposed "perceptual ruler" to trace each line stroke-by-stroke. Surprisingly, such a human-like perception manner enjoys an inference time scaling law -- the slower, the better. Researchers strive to speed up the model's perception in the past, but we slow it down again, allowing the model to read the image step-by-step and carefully.