MoCH: Moving Camouflaged Human Object Detection Dataset

Dataset Summary

MoCH is a dataset of Video Camouflaged Object Detection specific for human. It was created to support research in video camouflaged object detection..

The dataset contains:

• 98 video sequences and 4,091 human annotated frames.
• It has sampled frames and corrsponding image format annotations.
• Raw annotations are included.

Supported Tasks

This dataset supports the following tasks:

• Video Camouflaged Object Detection
• Video Camouflaged Human Detection
• Camouflaged Object Detection

Dataset Structure

Data Fields

Each video sequence in the dataset contains the following folders:

• images: the sampled original frames.
• gts: the binary mask/segmentation images for target camouflaged objects.
• 'annotations': raw annotation in coco format.

Data Splits

Split	# Videos
Train	49
Val	15
Test	34

Dataset Creation

Source Data

The data was collected from Youtube, and the original souce links of the videos in this dataset are available in this repo.

Annotations

The data was annotated by human annotators and annotations are also reviewd by human.

Licensing

The dataset is licensed under CC BY-NC 4.0. See [LICENSE] for details.