blenderproc.python.writer.CocoWriterUtility module

Allows rendering the content of the scene in the coco file format.

class blenderproc.python.writer.CocoWriterUtility._CocoWriterUtility

Bases: object

static bbox_from_binary_mask(binary_mask)

Returns the smallest bounding box containing all pixels marked “1” in the given image mask.

Parameters:

binary_mask (ndarray) – A binary image mask with the shape [H, W].

Return type:

List[int]

Returns:

The bounding box represented as [x, y, width, height]

static binary_mask_to_polygon(binary_mask, tolerance=0)

Converts a binary mask to COCO polygon representation

Parameters:

• binary_mask (ndarray) – a 2D binary numpy array where ‘1’s represent the object

• tolerance (int) – Maximum distance from original points of polygon to approximated polygonal chain. If tolerance is 0, the original coordinate array is returned.

Return type:

List[ndarray]

static calc_binary_mask_area(binary_mask)

Returns the area of the given binary mask which is defined as the number of 1s in the mask.

Parameters:

binary_mask (ndarray) – A binary image mask with the shape [H, W].

Return type:

int

Returns:

The computed area

static close_contour(contour)

Makes sure the given contour is closed.

Parameters:

contour (ndarray) – The contour to close.

Return type:

ndarray

Returns:

The closed contour.

static create_annotation_info(annotation_id, image_id, category_id, binary_mask, mask_encoding_format, tolerance=2)

Creates info section of coco annotation

Parameters:

• annotation_id (int) – integer to uniquly identify the annotation

• image_id (int) – integer to uniquly identify image

• category_id (int) – Id of the category

• binary_mask (ndarray) – A binary image mask of the object with the shape [H, W].

• mask_encoding_format (str) – Encoding format of the mask. Type: string.

• tolerance (int) – The tolerance for fitting polygons to the objects mask.

Return type:

Optional[Dict[str, Union[str, int]]]

static create_image_info(image_id, file_name, image_size)

Creates image info section of coco annotation

Parameters:

• image_id (int) – integer to uniquly identify image

• file_name (str) – filename for image

• image_size (Tuple[int, int]) – The size of the image, given as [W, H]

Return type:

Dict[str, Union[str, int]]

static generate_coco_annotations(inst_segmaps, inst_attribute_maps, image_paths, supercategory, mask_encoding_format, existing_coco_annotations=None, label_mapping=None)

Generates coco annotations for images

Parameters:

• inst_segmaps – List of instance segmentation maps

• inst_attribute_maps – per-frame mappings with idx, class and optionally supercategory/bop_dataset_name

• image_paths – A list of paths which points to the rendered images.

• supercategory – name of the dataset/supercategory to filter for, e.g. a specific BOP dataset

• mask_encoding_format – Encoding format of the binary mask. Type: string.

• existing_coco_annotations – If given, the new coco annotations will be appended to the given coco annotations dict.

• label_mapping (Optional[LabelIdMapping]) – The label mapping which should be used to label the categories based on their ids. If None, is given then the name field in the csv files is used or - if not existing - the category id itself is used.

Returns:

dict containing coco annotations

static merge_coco_annotations(existing_coco_annotations, new_coco_annotations)

Merges the two given coco annotation dicts into one.

Currently, this requires both coco annotations to have the exact same categories/objects. The “images” and “annotations” sections are concatenated and respective ids are adjusted.

Parameters:

• existing_coco_annotations – A dict describing the first coco annotations.

• new_coco_annotations – A dict describing the second coco annotations.

Returns:

A dict containing the merged coco annotations.

blenderproc.python.writer.CocoWriterUtility.binary_mask_to_rle(binary_mask)

Converts a binary mask to COCOs run-length encoding (RLE) format. Instead of outputting a mask image, you give a list of start pixels and how many pixels after each of those starts are included in the mask. :type binary_mask: ndarray :param binary_mask: a 2D binary numpy array where ‘1’s represent the object :rtype: Dict[str, List[int]] :return: Mask in RLE format

blenderproc.python.writer.CocoWriterUtility.rle_to_binary_mask(rle)

Converts a COCOs run-length encoding (RLE) to binary mask. :type rle: Dict[str, List[int]] :param rle: Mask in RLE format :rtype: ndarray :return: a 2D binary numpy array where ‘1’s represent the object

blenderproc.python.writer.CocoWriterUtility.write_coco_annotations(output_dir, instance_segmaps, instance_attribute_maps, colors, color_file_format='PNG', mask_encoding_format='rle', supercategory='coco_annotations', append_to_existing_output=True, jpg_quality=95, label_mapping=None, file_prefix='', indent=None)

Writes coco annotations in the following steps: 1. Locate the seg images 2. Locate the rgb maps 3. Locate the seg mappings 4. Read color mappings 5. For each frame write the coco annotation

Parameters:

• output_dir (str) – Output directory to write the coco annotations

• instance_segmaps (List[ndarray]) – List of instance segmentation maps

• instance_attribute_maps (List[dict]) – per-frame mappings with idx, class and optionally supercategory/bop_dataset_name

• colors (List[ndarray]) – List of color images. Does not support stereo images, enter left and right inputs subsequently.

• color_file_format (str) – Format to save color images in

• mask_encoding_format (str) – Encoding format of the binary masks. Default: ‘rle’. Available: ‘rle’, ‘polygon’.

• supercategory (str) – name of the dataset/supercategory to filter for, e.g. a specific BOP dataset set by ‘bop_dataset_name’ or any loaded object with specified ‘cp_supercategory’

• append_to_existing_output (bool) – If true and if there is already a coco_annotations.json file in the output directory, the new coco annotations will be appended to the existing file. Also, the rgb images will be named such that there are no collisions.

• jpg_quality (int) – The desired quality level of the jpg encoding

• label_mapping (Optional[LabelIdMapping]) – The label mapping which should be used to label the categories based on their ids. If None, is given then the name field in the csv files is used or - if not existing - the category id itself is used.

• file_prefix (str) – Optional prefix for image file names

• indent (Union[int, str, None]) – If indent is a non-negative integer or string, then the annotation output will be pretty-printed with that indent level. An indent level of 0, negative, or “” will only insert newlines. None (the default) selects the most compact representation. Using a positive integer indent indents that many spaces per level. If indent is a string (such as “ “), that string is used to indent each level.