ayase 0.1.15

Modular media quality metrics toolkit

Ayase

Modular media quality metrics toolkit.

⚠️ Work in progress. Some features may not work as expected.

Overview

• 246 quality metrics across visual, temporal, audio, perceptual, and safety categories.
• Modular pipeline - modules compute raw values, downstream apps decide what to do with them.
• CLI and Python API.
• Profile-based pipeline configuration.

Installation

Core (no ML models, metadata/structural checks only):

pip install ayase

With ML-based quality metrics:

pip install ayase[ml]           # Everything
pip install ayase[v-perceptual] # CLIP, LPIPS, open-clip, timm
pip install ayase[v-iqa]        # PyIQA, BRISQUE, NIQE, scikit-video
pip install ayase[v-motion]     # RAFT optical flow, decord
pip install ayase[v-ocr]        # PaddleOCR text recognition
pip install ayase[v-face]       # MediaPipe face detection
pip install ayase[v-audio]      # librosa audio analysis

Development:

pip install ayase[dev]          # pytest, black, ruff, mypy
pip install ayase[all]          # Everything including dev + TUI

See MODELS.md for the complete inventory of all pretrained weights used by every module. Models are downloaded and cached automatically on first use via HuggingFace Hub, Torch Hub, and PyIQA.

Metrics

208 modules across image, video, and audio quality assessment. See METRICS.md for the full reference with config parameters.

Categories: visual quality (NR/FR), perceptual similarity, motion, temporal consistency, audio, text alignment, HDR, safety, face, depth, codec.

Quick Start

CLI

# Scan a dataset and get a report
ayase scan ./my_dataset

# Scan with specific modules
ayase scan ./my_dataset --modules metadata,basic_quality,motion

# List all available modules
ayase modules list

# Check which modules can be loaded (dependencies installed)
ayase modules check

# Filter dataset by quality score
ayase filter ./my_dataset --min-score 70 --output ./filtered

Python API (recommended)

from ayase import AyasePipeline

ayase = AyasePipeline(modules=["basic"])
results = ayase.run("./my_dataset")

for path, sample in results.items():
    if sample.quality_metrics:
        print(f"{sample.path.name}: technical={sample.quality_metrics.technical_score}")

print(f"Total: {ayase.stats.total_samples}, Valid: {ayase.stats.valid_samples}")
ayase.export("report.json")

AyasePipeline accepts three ways to configure modules:

# By module names
ayase = AyasePipeline(modules=["metadata", "basic_quality", "motion"])

# By profile dict
ayase = AyasePipeline(profile={
    "name": "my_check",
    "modules": ["basic", "aesthetic"],
    "module_config": {
        "aesthetic": {"model_name": "openai/clip-vit-large-patch14"},
    },
})

# By profile file
ayase = AyasePipeline(profile="my_profile.toml")

# With custom config
from ayase.config import AyaseConfig
ayase = AyasePipeline(config=AyaseConfig(general={"parallel_jobs": 16}), modules=["basic"])

Low-level Pipeline API

import asyncio
from pathlib import Path
from ayase.pipeline import Pipeline, ModuleRegistry
from ayase.scanner import scan_dataset

ModuleRegistry.discover_modules()
module_names = ["metadata", "basic_quality", "semantic_alignment"]
modules = [ModuleRegistry.get_module(n)() for n in module_names]

pipeline = Pipeline(modules)
pipeline.start()

samples = scan_dataset(Path("./my_dataset"), recursive=True)
for sample in samples:
    processed = asyncio.run(pipeline.process_sample(sample))

pipeline.stop()
pipeline.export_report("report.json", format="json")

Profile-based pipelines

from ayase import load_profile, instantiate_profile_modules

profile = load_profile("my_profile.toml")
modules = instantiate_profile_modules(profile)
# modules is a list of PipelineModule instances ready for Pipeline()

Configuration

Create ayase.toml in your project root:

[general]
parallel_jobs = 8
cache_enabled = true

[quality]
enable_blur_detection = true
blur_threshold = 100.0

[pipeline]
dataset_path = "./my_dataset"
modules = ["metadata", "basic_quality", "motion"]
plugin_folders = ["plugins"]

[output]
default_format = "markdown"
artifacts_dir = "reports"
artifacts_format = "json"

[filter]
default_mode = "list"
min_score_threshold = 60

Ayase looks for config in: ./ayase.toml -> ~/.config/ayase/config.toml -> built-in defaults.

Writing Plugins

Create a .py file in your plugins/ folder:

from ayase.models import Sample, ValidationIssue, ValidationSeverity
from ayase.pipeline import PipelineModule

class MyCustomCheck(PipelineModule):
    name = "my_check"
    description = "Custom quality check"
    default_config = {"threshold": 0.5}

    def process(self, sample: Sample) -> Sample:
        # Your logic here
        if some_score < self.config["threshold"]:
            sample.validation_issues.append(
                ValidationIssue(
                    severity=ValidationSeverity.WARNING,
                    message="Quality below threshold",
                )
            )
        return sample

Then run:

ayase scan ./data --modules metadata,my_check

Development

git clone <repo-url>
cd ayase
pip install -e ".[dev]"

# Run tests
pytest

# Lint and format
ruff check src/ tests/
black src/ tests/

# Type check
mypy src/ayase

License

MIT - see LICENSE.