tinyperson
收藏资源简介:
随着深度卷积神经网络的兴起,视觉目标检测取得了前所未有的进展。然而,在大尺度图像中检测小于20像素的极小目标仍然没有得到很好的研究。对于极小目标的检测,一方面的挑战来自于其特征表示微弱,另一方面是复杂背景中存在大量相似特征增加了误报的风险。为了促进对于极小目标检测的研究,Yu等提出该数据集——TinyPerson,这是第1个远距离和大背景下进行人员检测的基准,为极小目标检测开辟了一个新的前景方向。
With the rise of deep convolutional neural networks, visual object detection has achieved unprecedented progress. However, detecting tiny objects smaller than 20 pixels in large-scale images remains understudied. For tiny object detection, on one hand, the challenge stems from their weak feature representation; on the other hand, the abundance of similar features in complex backgrounds elevates the risk of false positives. To advance research on tiny object detection, Yu et al. proposed this dataset—TinyPerson, which is the first benchmark for person detection under long-distance and large background scenarios, opening a new promising direction for tiny object detection research.




