MoCA-Mask

我们根据移动伪装动物 (MoCA) 策划了一个大规模的VCOD数据集。新的数据集，或简称MoCA掩码，包含87个视频序列，总共22,939帧，具有像素级地面真相掩码。MoCA-Mask封装了各种挑战，例如复杂的背景以及小巧而伪装的物体。我们为数据集中的所有视频提供每五帧的注释，边界框和密集分割蒙版。我们还为现有的VCOD方法提供了第一个综合基准。

We curated a large-scale Video Camouflaged Object Detection (VCOD) dataset based on Mobile Camouflaged Animals (MoCA). The new dataset, abbreviated as MoCA-Mask, consists of 87 video sequences totaling 22,939 frames with pixel-level ground-truth masks. MoCA-Mask encompasses various challenging scenarios, including complex backgrounds and small, well-camouflaged objects. We provide annotations for every fifth frame across all videos in the dataset, including bounding boxes and dense segmentation masks. Additionally, we present the first comprehensive benchmark for existing VCOD methods.

AI搜集汇总

背景与挑战

背景概述

MoCA-Mask是一个大规模视频伪装物体检测(VCOD)数据集，包含87个视频序列共22,939帧，提供像素级地面真相掩码。该数据集专注于移动伪装动物，具有复杂背景和小巧伪装物体等挑战，为每五帧提供注释、边界框和密集分割蒙版，并建立了首个VCOD方法的综合基准。