SyB3R - Synthetic Benchmark for 3D Reconstruction
|
SyB3R comes with a small set of 3D models, image data, and ground truth. These dataset have been generated to illustrate the potential use of SyB3R and are not meant for a rigorous evaluation of any part of the 3D reconstruction pipeline. Nevertheless, the 3D models contain several challenging aspects (such as reflective or homogeneous surfaces, areas of low texture, self-occlusion, etc.) while the rendered images are deteriorated with realistic noise, camera motion blur, depth of field, etc.
You can use these models and images in order to run tests of your 3D reconstruction pipeline. This section describes how to generate own image sequences and reference data in case, you prefer to use your own benchmarks. if its cool, share it, eg. send it to use so we can upload it to SyB3R webpage
Open blender, file, open, select path of the model file important short cuts: select item: right click apply action: left click move item: g rotate item: r change views: axis-aligned view (7,1,3), gradual change (8,4,2) change viewport shading: click button at the bottom of viewer <add image>
add, camera rotation and translation as above change to camera view: 0
file --> user preferences --> addons --> install from file select SyB3R/blenderUtils/Syb3rBlenderTools.py select category: ``Syb3r'', activate addon if permanent is wanted: save user settings
file –> initizes scene for blender
add recommendated settings
add recommendated settings
add recommendated settings
save scene in blender file --> save as
how?
Syb3r/build/executables/SynthesizeDataset/synthesizeDataset Usage: ./synthesizeDataset --filename renderScript.py --outputDir /path/to/output/directory/ [--camera cameraLeft] [--camera cameraRight] [--GPU] [--preRenderHook preRenderHookScript.py]