Copyright (c) 2023-2024 Antmicro
This repository provides instance segmentation annotations in the COCO JSON format for a subset of the Lindenthal Camera Traps dataset.
The Lindenthal Camera Traps Dataset comprises 775 video recordings captured at the Lindenthal Wildlife Park in Cologne, Germany, using an Intel RealSense D435 stereo camera.
The annotations were created by excluding the first 20 frames of each video to avoid any influence from the Intel RealSense D435 stereo camera's automatic exposure adjustment. From the remaining frames, every 10th frame was annotated.
Each annotation includes:
- Instance Mask: Pixel-level segmentation for precise shape definition of each animal.
- Bounding Box: Rectangular regions enclosing each detected animal, useful for object detection tasks.
- Class Label: Specific identification of the animal species.
The annotations are provided in the annotations_lindenthal.json file and are licensed under the Apache-2.0 license.
The video recordings are available for download from the LILA repository and are subject to the licensing terms specified on the website.