Skip to content

Latest commit

 

History

History

generate_image_dataset

README.md

Generate Image Dataset

In this example application, we demonstrate how to generate an image dataset using our freeform camera agent.

Before running this example, rename user_config.yaml.example to user_config.yaml and modify the contents appropriately for your system, as described in our Getting Started tutorial.

Important configuration options

You can control the behavior of this example by setting the following parameters in your user_config.yaml file, e.g.,

  • SPEAR.PAKS_DIR is the directory containing scene data in the form of PAK files.
  • SP_SERVICES.LEGACY.CAMERA_AGENT.CAMERA.RENDER_PASSES can be set to a list of image modalities that you want the agent to return (e.g., setting the value ["depth", "final_color", "segmentation"] will return depth images, photorealistic RGB images, and segmentation images).

Your user_config.yaml file only needs to specify the value of a parameter if it differs from the defaults defined in the python/config directory. You can browse this directory for a complete set of all user-configurable parameters.

Running the example

You can run the example as follows.

# generate camera poses
python generate_poses.py

# generate images
python generate_images.py

Running generate_poses.py will generate a poses.csv file consisting of camera poses that will be used in the following step. This tool accepts several optional command-line arguments that can be used to control its behavior (see the source code for details).

Running generate_images.py will generate images in an images directory. This tool accepts several optional command-line arguments that can be used to control its behavior (see the source code for details), e.g.,

  • --poses_file can be used to generate images based on the camera poses in a specific CSV file.
  • --num_internal_steps can be used to control overall image quality, since Unreal aggregates rendering information across multiple frames to compute various rendering effects.
  • --benchmark can be used to test the overall speed of the simulation.
  • --wait_for_key_press can be used to compare the game window output to the image that has been saved to disk.