Known issues
A major focus of this dataset is to help improve methods for estimating 3D humans from video. The dataset is also useful for training and evaluating methods for 3D/4D point tracking, structure from motion estimation with non-rigid motions, depth estimation, optical flow, and dynamic scene reconstruction. The following list of known dataset issues can help you decide if the dataset quality is sufficient for your use case.
Rendering
Environment specific issues:
- BE_CitySample: non flat ground plane at some locations
- Some of the rendered scenes include a ground plane with irregular height at city street and street intersection locations. Since the human motions used for B2 were captured on flat surfaces, this mismatch can cause the feet to occasionally interpenetrate the ground or appear to float above it. The following is a non-exhaustive list of scenes affected by this issue, which may be updated over time:
- Shots on sidewalks have flat ground plane and are not affected
Clothing simulations
The document linked below lists clothing simulations, each assigned a realism score from 0 to 6. These scores were manually determined by reviewing the simulations to exclude particularly bad cases of interpenetration or jittering. Simulations with a score below 3 were discarded. For applications requiring high-quality clothing simulations, the remaining simulations can be further filtered using these realism scores.