Accurate Gigapixel Crowd Counting by Iterative Zooming and Refinement
This addresses the challenge of accurate crowd counting in high-resolution gigapixel images for applications like surveillance and event management, representing a novel method for a known bottleneck rather than an incremental improvement.
The paper tackles the problem of crowd counting in gigapixel images, which exceed GPU memory and computation limits, by proposing GigaZoom, a method that iteratively zooms into dense areas and refines density maps, achieving state-of-the-art results with a 42% improvement in accuracy over the next best method.
The increasing prevalence of gigapixel resolutions has presented new challenges for crowd counting. Such resolutions are far beyond the memory and computation limits of current GPUs, and available deep neural network architectures and training procedures are not designed for such massive inputs. Although several methods have been proposed to address these challenges, they are either limited to downsampling the input image to a small size, or borrowing from other gigapixel tasks, which are not tailored for crowd counting. In this paper, we propose a novel method called GigaZoom, which iteratively zooms into the densest areas of the image and refines coarser density maps with finer details. Through experiments, we show that GigaZoom obtains the state-of-the-art for gigapixel crowd counting and improves the accuracy of the next best method by 42%.