MDBD (Multicue Dataset for Edge Detection)

为了研究在具有挑战性的自然场景中边界检测过程中几个早期视觉线索（亮度、颜色、立体、运动）的相互作用，我们使用消费者构建了一个由自然场景的短双目视频序列组成的多线索视频数据集-级富士立体相机（Mély、Kim、McGill、Guo 和 Serre，2016 年）。我们考虑了各种地点（从大学校园到街景和公园）和季节，以尽量减少可能的偏见。我们试图通过在每个镜头中以各种外观取景来捕捉更具挑战性的场景以进行边界检测。代表性示例关键帧如下图所示。该数据集包含 100 个场景，每个场景由左右视图短（10 帧）颜色序列组成。每个序列以每秒 30 帧的速率进行采样。每帧的分辨率为 1280 x 720 像素。

To investigate the interplay of several early visual cues (brightness, color, stereo, and motion) during boundary detection in challenging natural scenes, we constructed a multi-cue video dataset composed of short binocular video sequences of natural scenes using a consumer-grade Fujifilm stereo camera (Mély, Kim, McGill, Guo, and Serre, 2016). We considered diverse locations ranging from university campuses, street scenes to parks, as well as different seasons, to minimize potential biases. We aimed to capture more challenging boundary detection scenarios by framing each shot with varied visual appearances. Representative key frames of sample sequences are shown below. This dataset comprises 100 scenes, each consisting of short (10-frame) color sequences for left and right views. Each sequence is sampled at 30 frames per second, and each frame has a resolution of 1280 × 720 pixels.

搜集汇总

背景与挑战

背景概述

MDBD是一个用于边缘检测的多线索视频数据集，旨在研究自然场景中亮度、颜色、立体和运动等早期视觉线索的相互作用。该数据集包含100个场景，每个场景由左右视图的10帧颜色序列组成，分辨率为1280 x 720像素，帧率为30fps，采集自多样化的地点和季节以减少偏见，适用于边界检测算法的开发和评估。