Implementation:Datajuicer Data juicer ImageRemoveBackgroundMapper

Knowledge Sources	Datajuicer_Data_juicer
Domains	Data_Processing, Mapping
Last Updated	2026-02-14 16:00 GMT

Overview

Concrete tool for removing backgrounds from images in the dataset provided by Data-Juicer.

Description

ImageRemoveBackgroundMapper is a mapper operator that removes the background from images using the rembg library (backed by ONNX Runtime). It supports optional alpha matting with configurable foreground/background thresholds and erosion size for smoother edges. A custom background color can be specified via the bgcolor parameter. Processed images are saved in PNG format to a configurable output directory, and source file paths in the sample are updated accordingly.

Usage

Use when you need to isolate foreground subjects from their backgrounds for creating clean subject-focused training data, image segmentation datasets, and compositing workflows.

Code Reference

Source Location

• Repository: Datajuicer_Data_juicer
• File: data_juicer/ops/mapper/image_remove_background_mapper.py

Signature

@OPERATORS.register_module("image_remove_background_mapper")
class ImageRemoveBackgroundMapper(Mapper):
    def __init__(self,
                 alpha_matting: bool = False,
                 alpha_matting_foreground_threshold: int = 240,
                 alpha_matting_background_threshold: int = 10,
                 alpha_matting_erode_size: int = 10,
                 bgcolor: Optional[Tuple[int, int, int, int]] = None,
                 save_dir: str = None,
                 *args, **kwargs):

Import

from data_juicer.ops.mapper.image_remove_background_mapper import ImageRemoveBackgroundMapper

I/O Contract

Inputs

Name	Type	Required	Description
alpha_matting	bool	No	Whether to use alpha matting, defaults to False
alpha_matting_foreground_threshold	int	No	Foreground threshold for alpha matting, defaults to 240
alpha_matting_background_threshold	int	No	Background threshold for alpha matting, defaults to 10
alpha_matting_erode_size	int	No	Erosion size for alpha matting, defaults to 10
bgcolor	Optional[Tuple[int, int, int, int]]	No	Background color for the cutout image (RGBA), defaults to None
save_dir	str	No	Directory to store generated images; if not specified, saves in same directory as input

Outputs

Name	Type	Description
samples	Dict	Transformed samples with background-removed image paths updated

Usage Examples

process:
  - image_remove_background_mapper:
      alpha_matting: false

Related Pages

• Environment:Datajuicer_Data_juicer_Python_Runtime_Environment