Skip to content

Latest commit

 

History

History
56 lines (39 loc) · 3.27 KB

Running_generation.md

File metadata and controls

56 lines (39 loc) · 3.27 KB
Raw

Running dataset generation

Before running generation, make sure that you are familiar with setup and tested your setup with GarmentViever GUI.

The process consists of three steps that need to be run sequentially:

  • Sampling sewing pattern designs from template
  • Draping each over the base body (physics simulation)
  • (if needed) perform 3D scan imitation on each of the meshes

More on each step below.

Components of the generation system

Run help(module) for even more detailed descriptions of corresponding python modules

Dataset generator

In datasetgenerator.py module

2D pattern dataset generation from given templates. Allows to configure the generation by supplying Properties object. Example usage of the generator is given in if __name__ == "__main__": section of the file.

Simulation

datasim.py & mayaqltools package

  • Simulates each pattern in the dataset over a given human body object
  • Uses Qualoth for Maya cloth simulation
  • It is designed to run with Mayapy as standalone interpreter
  • Command line arguments are the dirs\files basenames given that system.json is properly created.

Example usage from command line:

<Maya Installation path>/bin/mayapy.exe "./datasim.py" --data <dataset_name> --minibatch <size>  --config <simulation_props.json>

Running simulation of large-scale datasets over Maya\Qualoth crashes

dataset_runner.sh script is given for convence of processing large amounts of garment patters over long period of time. The main feature is detection of dataset sim processing hangs \ crashes and automatic resume of dataset processing in case of such events.

Imitating 3D scanning artifacts

datascan.py

  • is designed to work on datasets that passed the simulation steps.
  • It modifies every 3D garment mesh to imitate missing geometry due to 3D scanning, and saves it as a separate file in the datapoint folder
  • The process does NOT override the initial meshes, but rather created new .obj files for corrupted versions.

Optional (but helpful) utilities

Found in utility scripts/

  • gather_renders.py is a small skript to copy all the simulation renders of each datapoint to one location for convenience of data review.
  • all_data_has_all_files.py tests if all datapoints in all datasets of your dataset folder are present and correctly structured (recommended to use after downloading the data or merging the datasets).
  • maya_segmentaion_viz.py a script to be executed within Maya environment to visualize segmentation of a mesh from a particular datapoint.
  • merge_datasets.py merges two dataset folders that were produced from the same template into one data folder with single dataset_properties.json file. It's helpful to keep the data organized by garment type.
  • crashes_to_unprocessed.py small utility for cases when the simulation process of dataset produced a lot of crashed examples and those need to be re-simulating without revising the correct ones.