Estimating camera motion in deformable scenes poses a complex and open research challenge. Most existing non-rigid structure from motion techniques assume to observe also static scene parts besides deforming scene parts in order to establish an anchoring reference. However, this assumption does not hold true in certain relevant application cases such as endoscopies. To tackle this issue with a common benchmark, we introduce the Drunkard’s Dataset, a challenging collection of synthetic data targeting visual navigation and reconstruction in deformable environments. This dataset is the first large set of exploratory camera trajectories with ground truth inside 3D scenes where every surface exhibits non-rigid deformations over time. Simulations in realistic 3D buildings lets us obtain a vast amount of data and ground truth labels, including camera poses, RGB images and depth, optical flow and normal maps at high resolution and quality.
1 PAPER • 1 BENCHMARK
The dataset includes the synthetic data generated from rendering the 3D meshes of LM objects and several household objects in Blender for training 6D pose estimation algorithms. The whole dataset contains synthetic data for 18 objects (13 from LM and 5 from household objects), with 20,000 data samples for each object. Each data sample includes an RGB image in .png format and a depth image in .exr format. Each sample has the annotations of mask labels in .png format and the ground truth pose labels saved in .json files. Apart from the training data, the 3D meshes of the objects and the pre-trained models of the 6D pose estimation algorithm are also included. The whole dataset takes approximately ~1T of storage memory.
1 PAPER • NO BENCHMARKS YET