suncg_with_cam_sampling

SUNCG scene with custom camera sampling

In contrast to the SUNCG basic example, we do here not load precomputed camera poses, but sample them using the SuncgCameraSampler module.

Usage

Execute in the BlenderProc main directory:

python run.py examples/suncg_with_cam_sampling/config.yaml <path to house.json> examples/suncg_with_cam_sampling/output

• examples/suncg_with_cam_sampling/config.yaml: path to the configuration file with pipeline configuration.
• <path to house.json>: Path to the house.json file of the SUNCG scene you want to render.
• examples/suncg_with_cam_sampling/output: path to the output directory.

Steps

• Loads a SUNCG scene: loader.SuncgLoader module.
• Sample camera positions inside every room: camera.SuncgCameraSampler module.
• Automatically adds light sources inside each room: lighting.SuncgLighting module.
• Writes sampled camera poses to file: writer.CameraStateWriter module.
• Renders normals: renderer.NormalRenderer module.
• Renders semantic segmentation map: renderer.SegMapRenderer module.
• Renders rgb and depth: renderer.RgbRenderer module.
• Merges all into an .hdf5 file: writer.Hdf5Writer module.

Config file

SuncgCameraSampler

{
  "module": "camera.SuncgCameraSampler",
  "config": {
    "cam_poses": [{
      "number_of_samples": 10,
      "proximity_checks": {
        "min": 1.0
      },
      "min_interest_score": 0.4,
      "location": {
        "provider":"sampler.Uniform3d",
        "max":[0, 0, 2],
        "min":[0, 0, 0.5]
      },
      "rotation": {
        "value": {
          "provider":"sampler.Uniform3d",
          "max":[1.2217, 0, 6.283185307],
          "min":[1.2217, 0, 0]
        }
      },
    }]
  }

With this module we want to sample 10 valid camera poses inside the loaded SUNCG rooms. The x and y coordinate are hereby automatically sampled uniformly across a random room, while we configure the z coordinate to lie between 0.5m and 2m above the ground. Regarding the camera rotation we fix the pitch angle to 70°, the roll angle to 0° and sample the yaw angle uniformly between 0° and 360°.

After sampling a pose the pose is only accepted if it is valid according to the properties we have specified:

• Per default a camera pose is only accepted, if there is no object between it and the floor
• As we enabled proximity_checks with a min value of 1.0, we then only accept the pose if every object in front of it is at least 1 meter away
• At the end we also check if the sampled view is interesting enough. Therefore a score is calculated based on the number of objects that are visible and how much space they occupy. Only if the score is above 0.4 the pose is accepted.

Visualization

Visualize the generated data:

python scripts/visHdf5Files.py examples/suncg_with_cam_sampling/output/0.hdf5

More examples

• sung_basic: More on rendering SUNCG scenes with fixed camera poses.