Deep Learning in Computer Vision by Hassaballah Mahmoud; Awad Ali Ismail;

Author:Hassaballah, Mahmoud; Awad, Ali Ismail;
Language: eng
Format: epub
Publisher: CRC Press LLC

6.3.2.2 Synthetic Datasets

Virtual KITTI [61] consists of 50 high-resolution videos consisting of 21,260 image frames of 1,242 × 375 resolution, generated using a unity game engine in an urban environment. Different annotations are provided with the dataset like semantic and instance segmentation, dense optical flow (DOF), depth maps, and 2D and 3D object detection bounding boxes. The depth map is a PNG16 image of the same size of the input image. The value of each pixel represents the z coordinate of the point in camera coordinate space. Each pixel takes values from 0 to 65,535, which corresponds to 655.36 m from the camera image plane. Points at infinity, like sky, are clipped to a depth of 655.3 m. The DOF image is a representation of dense optical flow between two consecutive images. Semantic and instance information is given per frame as a unique color for each class. Fourteen classes are annotated: Building, Car, GuardRail, Misc, Pole, Road, Sky, Terrain, TrafficLight, TrafficSign, Tree, Truck, Van, and Vegetation.

SYNTHIA [62] consists of 200,000 images from video sequences with a wide range of diversity regarding weather conditions and scene structure. 360° views are provided by simulating 8 RGB cameras. Depth maps are provided as well.

NUSCENES [63] is a large-scale dataset providing the most sensor measurements. It contains 1,000 videos of 20 seconds each with several types of annotation data such as LiDAR, radar, camera, IMU, and GPS data. It also provides 3D bounding boxes over 25 classes of objects annotated at 2 Hz.

SYNSCAPES [64] is very close to reality and provides both 1,440 × 720 and 2,048 × 1,024 resolution images, depth maps, semantic and instance segmentation labels. 2D and 3D bounding boxes are also provided. Each image of the 25,000 images has its own unique structure, which gives a very wide range of variations and features combinations.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.