Implementation:Datajuicer Data juicer ImageAestheticsFilter

Knowledge Sources	Datajuicer_Data_juicer
Domains	Data_Quality, Filtering
Last Updated	2026-02-14 16:00 GMT

Overview

Concrete tool for filtering data samples based on image aesthetics scores provided by Data-Juicer.

Description

ImageAestheticsFilter is a filter operator that keeps samples with aesthetics scores within a specific range. It uses a HuggingFace model (default: shunk031/aesthetics-predictor-v2-sac-logos-ava1-l14-linearMSE) to predict image aesthetics. Scores are normalized by dividing by 10 if the model name includes 'shunk031/aesthetics-predictor'. The operator supports CUDA acceleration and 'any'/'all' strategies across multiple images per sample. The key metric image_aesthetics_scores is cached in the stats field. It extends the Filter base class and implements the two-phase compute_stats/process pattern.

Usage

Import this operator when you need to filter dataset samples based on the aesthetic quality of images. Configure it in your Data-Juicer YAML config or instantiate directly.

Code Reference

Source Location

• Repository: Datajuicer_Data_juicer
• File: data_juicer/ops/filter/image_aesthetics_filter.py
• Lines: 1-128

Signature

@OPERATORS.register_module("image_aesthetics_filter")
@LOADED_IMAGES.register_module("image_aesthetics_filter")
class ImageAestheticsFilter(Filter):
    def __init__(
        self,
        hf_scorer_model: str = "",
        trust_remote_code: bool = False,
        min_score: float = 0.5,
        max_score: float = 1.0,
        any_or_all: str = "any",
        *args,
        **kwargs,
    ):
        ...

Import

from data_juicer.ops.filter.image_aesthetics_filter import ImageAestheticsFilter

I/O Contract

Inputs

Name	Type	Required	Description
hf_scorer_model	str	No	HuggingFace model name for aesthetics prediction. Default: "shunk031/aesthetics-predictor-v2-sac-logos-ava1-l14-linearMSE"
trust_remote_code	bool	No	Whether to trust remote code of HF models. Default: False
min_score	float	No	Minimum aesthetics score to keep samples. Default: 0.5
max_score	float	No	Maximum aesthetics score to keep samples. Default: 1.0
any_or_all	str	No	Keep strategy: 'any' or 'all' across images. Default: "any"

Outputs

Name	Type	Description
samples	Dict	Filtered samples with stats field updated (image_aesthetics_scores)

Usage Examples

YAML Configuration

process:
  - image_aesthetics_filter:
      min_score: 0.5
      max_score: 1.0
      any_or_all: "any"

Python API

from data_juicer.ops.filter.image_aesthetics_filter import ImageAestheticsFilter

op = ImageAestheticsFilter(min_score=0.5, max_score=1.0)
# Apply to dataset
result = dataset.process(op)

Related Pages

• Environment:Datajuicer_Data_juicer_Python_Runtime_Environment