Skip to content

Latest commit

 

History

History
65 lines (45 loc) · 2.65 KB

File metadata and controls

65 lines (45 loc) · 2.65 KB

COCO annotations

The focus of this example is to introduce user to writer.CocoAnnotationsWriter module.

Usage

Execute in the BlenderProc main directory:

blenderproc run examples/advanced/coco_annotations/main.py examples/resources/camera_positions examples/advanced/coco_annotations/scene.blend examples/advanced/coco_annotations/output
  • examples/advanced/coco_annotations/main.py: path to the main python file to run.
  • examples/resources/camera_positions: text file with parameters of camera positions.
  • examples/advanced/coco_annotations/scene.blend: path to the blend file with the basic scene.
  • examples/advanced/coco_annotations/output: path to the output directory.

Visualizing Annotations

You can use vis coco_annotation.py with the following command to visualize the annotations blended over a rendered rgb image:

blenderproc vis coco [-i <image index>] [-c <coco annotations json>] [-b <base folder of coco json and image files>]

Working Examples

With specific values:

blenderproc vis coco -i 1 -c coco_annotations.json -b examples/advanced/coco_annotations/output/coco_data

Above are also the default values, i.e. for the same result call:

blenderproc vis coco

Implementation

# Set some category ids for loaded objects
for j, obj in enumerate(objs):
    obj.set_cp("category_id", j+1)

# Render segmentation data and produce instance attribute maps
seg_data = bproc.renderer.render_segmap(map_by=["instance", "class", "name"])

To create coco annotations we need to render both instance and class maps. The class is defined in terms of a custom property category_id which must be previously defined for each instance. The category_id can be either set by a custom property as above or in a loader or can be directly defined in a .blend file. We also add "name" to the mapping, s.t. we can later use the object's names for labeling the categories in the coco annotations writer.

# Write data to coco file
bproc.writer.write_coco_annotations(os.path.join(args.output_dir, 'coco_data'),
                                    instance_segmaps=seg_data["instance_segmaps"],
                                    instance_attribute_maps=seg_data["instance_attribute_maps"],
                                    colors=data["colors"],
                                    color_file_format="JPEG")

This function stores annotations in coco_annotations.json. Optionally, you can set "supercategory": "<some_supercategory>" in the writer.CocoAnnotationsWriter config to filter objects by a previously assigned custom property "supercategory".