ayase 0.1.14

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

Input: img/vid = image or video, +ref = needs reference_path, +cap = needs caption, batch = dataset-level.

#	Metric	Module	Input	Description
1	blur_score	basic_quality	img/vid	Laplacian variance
2	compression_score	basic_quality	img/vid	Compression quality (0-100)
3	aesthetic_score	aesthetic	img/vid	0-10, from aesthetic predictor
4	clip_score	semantic_alignment	img/vid +cap	Caption-image alignment
5	brightness	basic_quality	img/vid	Mean brightness (0-255)
6	contrast	basic_quality	img/vid	Contrast level (0-100)
7	saturation	basic_quality	img/vid	Color saturation (0-100)
8	fast_vqa_score	fast_vqa	img/vid	0-100
9	motion_score	motion	img/vid	Scene motion intensity
10	camera_motion_score	stabilized_motion	img/vid	Camera motion intensity
11	temporal_consistency	clip_temporal	img/vid	Frame consistency
12	technical_score	basic_quality	img/vid	Composite technical score
13	noise_score	basic_quality	img/vid	Noise level (0-100, lower=cleaner)
14	artifacts_score	basic_quality	img/vid	Artifact severity (0-100, lower=better)
15	watermark_probability	watermark_classifier	img/vid	0-1
16	ocr_area_ratio	text_detection	img	0-1
17	face_count	face_fidelity	img/vid	Integer face count
18	nsfw_score	nsfw	img	0-1, likelihood of being NSFW
19	audio_quality_score	audio	audio	0-100
20	perceptual_hash	dedup	img/vid	dHash or similar
21	depth_score	depth_map_quality	img/vid	Scene depth complexity
22	auto_caption	captioning	img/vid +cap	Generated caption
23	vqa_a_score	aesthetic	img/vid	VQA aesthetic sub-score
24	vqa_t_score	basic_quality	img/vid	VQA technical sub-score
25	is_score	inception_score	img/vid	Inception Score
26	sd_score	sd_reference	img/vid +cap	SD-reference similarity (0-1)
27	gradient_detail	basic_quality	img/vid	Sobel gradient detail (0-100)
28	blip_bleu	captioning	img/vid +cap	BLIP caption BLEU score
29	detection_score	object_detection	img +cap	Object detection confidence
30	count_score	object_detection	img +cap	Object count accuracy
31	color_score	color_consistency	img/vid +cap	Color consistency
32	celebrity_id_score	celebrity_id	img/vid	Celebrity identity distance
33	identity_loss	identity_loss	img/vid +ref	Face identity distance (0-1, lower=better)
34	face_recognition_score	identity_loss	img/vid +ref	Face identity similarity (0-1, higher=better)
35	ocr_score	ocr_fidelity	img/vid +cap	OCR confidence
36	ocr_fidelity	ocr_fidelity	img/vid +cap	OCR text accuracy vs caption (0-100, higher=better)
37	ocr_cer	ocr_fidelity	img/vid +cap	Character Error Rate (0-1, lower=better)
38	ocr_wer	ocr_fidelity	img/vid +cap	Word Error Rate (0-1, lower=better)
39	i2v_clip	i2v_similarity	img/vid +ref	CLIP image-video similarity (0-1)
40	i2v_dino	i2v_similarity	img/vid +ref	DINOv2 image-video similarity (0-1)
41	i2v_lpips	i2v_similarity	img/vid +ref	LPIPS image-video distance (0-1, lower=better)
42	i2v_quality	i2v_similarity	img/vid +ref	Aggregated I2V quality (0-100)
43	action_score	action_recognition	img/vid +cap	Caption-action fidelity (0-100)
44	action_confidence	action_recognition	img/vid +cap	Top-1 action confidence (0-100)
45	flow_score	advanced_flow	img/vid	Optical flow magnitude
46	motion_ac_score	motion_amplitude	img/vid +cap	Motion amplitude
47	warping_error	temporal_flickering	img/vid	Warping error
48	clip_temp	clip_temporal	img/vid	CLIP temporal consistency
49	face_consistency	clip_temporal	img/vid	Face identity temporal consistency
50	psnr	structural	img/vid +ref	Peak Signal-to-Noise Ratio (dB, higher=better)
51	ssim	structural	img/vid +ref	Structural Similarity (0-1, higher=better)
52	lpips	perceptual_fr	img/vid +ref	Learned Perceptual distance (0-1, lower=better)
53	spectral_entropy	spectral_complexity	img/vid	DINOv2 spectral entropy
54	spectral_rank	spectral_complexity	img/vid	DINOv2 effective rank ratio
55	fvd	fvd	batch	Frechet Video Distance
56	kvd	fvd	batch	Kernel Video Distance
57	fvmd	fvmd	batch	Frechet Video Motion Distance
58	vmaf	vmaf	img/vid +ref	VMAF (0-100, higher=better)
59	ms_ssim	ms_ssim	img/vid +ref	Multi-Scale SSIM (0-1)
60	vif	vif	img +ref	Visual Information Fidelity
61	niqe	niqe	img	Natural Image Quality Evaluator (lower=better)
62	t2v_score	t2v_score	img/vid +cap	T2VScore alignment + quality
63	t2v_alignment	t2v_score	img/vid +cap	Text-video semantic alignment
64	t2v_quality	t2v_score	img/vid +cap	Video production quality
65	dynamics_range	dynamics_range	img/vid	Extent of content variation
66	dynamics_controllability	dynamics_controllability	img/vid +cap	Motion control fidelity
67	scene_complexity	scene_complexity	img/vid	Visual complexity score
68	compression_artifacts	compression_artifacts	img/vid	Artifact severity (0-100)
69	naturalness_score	naturalness	img	Natural scene statistics
70	video_memorability	video_memorability	img/vid	Memorability prediction
71	usability_rate	usability_rate	img	Percentage of usable frames
72	confidence_score	llm_descriptive_qa	img/vid	Prediction confidence
73	human_preference_score	llm_advisor	img/vid	Human preference (0-1, higher=better)
74	engagement_score	llm_advisor	img/vid	Engagement prediction (0-1, higher=better)
75	usability_score	usability_rate	img/vid	Usability estimate (0-100, higher=better)
76	hdr_quality	hdr_sdr_vqa	img/vid	HDR-specific quality
77	sdr_quality	hdr_sdr_vqa	img/vid	SDR-specific quality
78	temporal_information	ti_si	img/vid	ITU-T P.910 TI (higher=more motion)
79	spatial_information	ti_si	img/vid	ITU-T P.910 SI (higher=more detail)
80	flicker_score	flicker_detection	img/vid	Flicker severity 0-100 (lower=better)
81	judder_score	judder_stutter	img/vid	Judder severity 0-100 (lower=better)
82	stutter_score	judder_stutter	img/vid	Duplicate/dropped frames 0-100 (lower=better)
83	dists	dists	img/vid +ref	DISTS (0-1, lower=more similar)
84	fsim	perceptual_fr	img/vid +ref	Feature Similarity Index (0-1, higher=better)
85	gmsd	perceptual_fr	img/vid +ref	Gradient Magnitude Similarity Deviation (lower=better)
86	vsi_score	perceptual_fr	img/vid +ref	Visual Saliency Index (0-1, higher=better)
87	brisque	brisque	img/vid	BRISQUE (0-100, lower=better)
88	pesq_score	audio_pesq	audio +ref	PESQ (-0.5 to 4.5, higher=better)
89	estoi_score	audio_estoi	audio +ref	ESTOI speech intelligibility (0-1, higher=better)
90	mcd_score	audio_mcd	audio +ref	Mel Cepstral Distortion (dB, lower=better)
91	si_sdr_score	audio_si_sdr	audio +ref	Scale-Invariant SDR (dB, higher=better)
92	lpdist_score	audio_lpdist	audio +ref	Log-Power Spectral Distance (lower=better)
93	utmos_score	audio_utmos	audio	UTMOS predicted MOS (1-5, higher=better)
94	av_sync_offset	av_sync	audio	Audio-video sync offset in ms
90	dover_score	dover	img/vid	DOVER overall (higher=better)
91	dover_technical	dover	img/vid	DOVER technical quality
92	dover_aesthetic	dover	img/vid	DOVER aesthetic quality
93	topiq_score	topiq	img/vid	TOPIQ transformer-based IQA (higher=better)
94	liqe_score	liqe	img/vid	LIQE lightweight IQA (higher=better)
95	clip_iqa_score	clip_iqa	img/vid	CLIP-IQA semantic quality (0-1, higher=better)
96	color_grading_score	production_quality	img/vid	Colour consistency 0-100
97	white_balance_score	production_quality	img/vid	White balance accuracy 0-100
98	exposure_consistency	production_quality	img/vid	Exposure stability 0-100
99	focus_quality	production_quality	img/vid	Sharpness/focus quality 0-100
100	banding_severity	production_quality	img/vid	Colour banding 0-100 (lower=better)
101	qalign_quality	q_align	img/vid	Q-Align technical quality (1-5, higher=better)
102	qalign_aesthetic	q_align	img/vid	Q-Align aesthetic quality (1-5, higher=better)
103	face_quality_score	face_fidelity	img/vid	Composite face quality 0-100 (higher=better)
104	face_identity_consistency	face_landmark_quality	img/vid	Temporal face identity stability (0-1)
105	face_expression_smoothness	face_landmark_quality	img/vid	Expression smoothness (0-100)
106	face_landmark_jitter	face_landmark_quality	img/vid	Landmark jitter 0-100 (lower=better)
107	object_permanence_score	object_permanence	img/vid	Object tracking (0-100)
108	semantic_consistency	semantic_segmentation_consistency	img/vid	Segmentation IoU (0-1)
109	depth_temporal_consistency	depth_consistency	img/vid	Depth map correlation 0-1 (higher=better)
110	subject_consistency	subject_consistency	img/vid	Subject identity (0-1)
111	background_consistency	background_consistency	img/vid	Background stability (0-1, higher=better)
112	motion_smoothness	motion_smoothness	img/vid	Motion smoothness (0-1, higher=better)
113	codec_efficiency	codec_specific_quality	img/vid	Quality-per-bit efficiency 0-100 (higher=better)
114	gop_quality	codec_specific_quality	img/vid	GOP structure appropriateness 0-100 (higher=better)
115	codec_artifacts	codec_specific_quality	img/vid	Block artifact severity 0-100 (lower=better)
116	deepfake_probability	deepfake_detection	img/vid	Synthetic/deepfake likelihood 0-1
117	ai_generated_probability	watermark_classifier	img/vid	AI-generated content likelihood 0-1
118	harmful_content_score	harmful_content	img/vid	Violence/gore severity 0-1
119	watermark_strength	watermark_robustness	img/vid	Invisible watermark strength 0-1
120	bias_score	bias_detection	img/vid	Representation imbalance indicator 0-1
121	depth_quality	depth_map_quality	img/vid	Depth map quality 0-100 (higher=better)
122	multiview_consistency	multi_view_consistency	img/vid	Geometric consistency 0-1 (higher=better)
123	stereo_comfort_score	stereoscopic_quality	img/vid	Stereo viewing comfort 0-100 (higher=better)
124	musiq_score	musiq	img/vid	MUSIQ multi-scale IQA (higher=better)
125	contrique_score	contrique	img/vid	CONTRIQUE contrastive IQA (higher=better)
126	mdtvsfa_score	mdtvsfa	img/vid	MDTVSFA fragment-based VQA (higher=better)
127	nima_score	nima	img/vid	NIMA aesthetic+technical (1-10, higher=better)
128	dbcnn_score	dbcnn	img/vid	DBCNN bilinear CNN (higher=better)
129	wadiqam_score	wadiqam	img/vid	WaDIQaM-NR (higher=better)
130	maniqa_score	maniqa	img/vid	MANIQA multi-attention (higher=better)
131	arniqa_score	arniqa	img/vid	ARNIQA (higher=better)
132	qualiclip_score	qualiclip	img/vid	QualiCLIP opinion-unaware (higher=better)
133	pieapp	pieapp	img/vid +ref	PieAPP pairwise preference (lower=better)
134	cw_ssim	cw_ssim	img/vid +ref	Complex Wavelet SSIM (0-1, higher=better)
135	nlpd	nlpd	img/vid +ref	Normalized Laplacian Pyramid Distance (lower=better)
136	mad	mad	img/vid +ref	Most Apparent Distortion (lower=better)
137	ahiq	ahiq	img/vid +ref	Attention Hybrid IQA (higher=better)
138	topiq_fr	topiq_fr	img/vid +ref	TOPIQ full-reference (higher=better)
139	dreamsim	dreamsim	img/vid +ref	DreamSim CLIP+DINO similarity (lower=more similar)
140	cover_score	cover	img/vid	COVER overall (higher=better)
141	cover_technical	cover	img/vid	COVER technical branch
142	cover_aesthetic	cover	img/vid	COVER aesthetic branch
143	cover_semantic	cover	img/vid	COVER semantic branch
144	vqa_score_alignment	vqa_score	img/vid +cap	VQAScore text-visual alignment (0-1, higher=better)
145	videoscore_visual	videoscore	img/vid +cap	VideoScore visual quality
146	videoscore_temporal	videoscore	img/vid +cap	VideoScore temporal consistency
147	videoscore_dynamic	videoscore	img/vid +cap	VideoScore dynamic degree
148	videoscore_alignment	videoscore	img/vid +cap	VideoScore text-video alignment
149	videoscore_factual	videoscore	img/vid +cap	VideoScore factual consistency
150	face_iqa_score	face_iqa	img/vid	TOPIQ-face face quality (higher=better)
151	scene_stability	scene_detection	img/vid	Scene stability (0-1, 1=single continuous scene)
152	avg_scene_duration	scene_detection	img/vid	Average scene duration in seconds
153	raft_motion_score	raft_motion	img/vid	RAFT optical flow magnitude
154	ram_tags	ram_tagging	img/vid	Comma-separated RAM auto-tags
155	depth_anything_score	depth_anything	img/vid	Monocular depth quality
156	depth_anything_consistency	depth_anything	img/vid	Temporal depth consistency
157	video_type	video_type_classifier	img/vid	Content type (real, animated, game, etc.)
158	video_type_confidence	video_type_classifier	img/vid	Classification confidence
159	jedi	jedi_metric	batch	Per-sample V-JEPA feature (batch-computed)
160	trajan_score	trajan	img/vid	Point track motion consistency
161	promptiqa_score	promptiqa	img/vid	Few-shot NR-IQA score
162	aigv_static	aigv_assessor	img/vid +cap	AI video static quality
163	aigv_temporal	aigv_assessor	img/vid +cap	AI video temporal smoothness
164	aigv_dynamic	aigv_assessor	img/vid +cap	AI video dynamic degree
165	aigv_alignment	aigv_assessor	img/vid +cap	AI video text-video alignment
166	video_reward_score	video_reward	img/vid +cap	Human preference reward
167	tifa_score	tifa	img/vid +cap	VQA faithfulness (0-1, higher=better)
168	text_overlay_score	text_detection	img	Text overlay severity (0-1)
169	ptlflow_motion_score	ptlflow_motion	img/vid	ptlflow optical flow magnitude
170	qcn_score	qcn	img/vid	Geometric order blind IQA
171	finevq_score	finevq	img/vid	FineVQ fine-grained UGC VQA (CVPR 2025)
172	kvq_score	kvq	img/vid	KVQ saliency-guided VQA (CVPR 2025)
173	rqvqa_score	rqvqa	img/vid	RQ-VQA rich quality-aware (CVPR 2024 winner)
174	videval_score	videval	img/vid	VIDEVAL 60-feature fusion NR-VQA
175	tlvqm_score	tlvqm	img/vid	TLVQM two-level video quality
176	funque_score	funque	img/vid +ref	FUNQUE unified quality (beats VMAF)
177	movie_score	movie	img/vid +ref	MOVIE motion trajectory FR
178	st_greed_score	st_greed	img/vid +ref	ST-GREED variable frame rate FR
179	c3dvqa_score	c3dvqa	img/vid	C3DVQA 3D CNN spatiotemporal FR
180	flolpips	flolpips	img/vid	FloLPIPS flow-based perceptual FR
181	hdr_vqm	hdr_vqm	img/vid +ref	HDR-VQM HDR video quality FR
182	st_lpips	st_lpips	img/vid	ST-LPIPS spatiotemporal perceptual FR
183	camera_jitter_score	camera_jitter	img/vid	Camera stability (0-1, 1=stable)
184	jump_cut_score	jump_cut	img/vid	Jump cut absence (0-1, 1=no cuts)
185	playback_speed_score	playback_speed	img/vid	Normal speed (1.0=normal)
186	flow_coherence	flow_coherence	img/vid	Bidirectional optical flow consistency (0-1)
187	letterbox_ratio	letterbox	img/vid	Border/letterbox fraction (0-1, 0=no borders)
188	tonal_dynamic_range	tonal_dynamic_range	img/vid	Luminance histogram span (0-100)
189	vtss	vtss	img	Video Training Suitability Score (0-1)
190	cnniqa_score	cnniqa	img/vid	CNNIQA blind CNN IQA
191	hyperiqa_score	hyperiqa	img/vid	HyperIQA adaptive NR-IQA
192	paq2piq_score	paq2piq	img/vid	PaQ-2-PiQ patch-to-picture (CVPR 2020)
193	tres_score	tres	img/vid	TReS transformer IQA (WACV 2022)
194	unique_score	unique	img/vid	UNIQUE unified NR-IQA (TIP 2021)
195	laion_aesthetic	laion_aesthetic	img/vid	LAION Aesthetics V2 (0-10)
196	compare2score	compare2score	img/vid	Compare2Score comparison-based
197	afine_score	afine	img/vid	A-FINE fidelity-naturalness (CVPR 2025)
198	ckdn_score	ckdn	img +ref	CKDN knowledge distillation FR
199	deepwsd_score	deepwsd	img +ref	DeepWSD Wasserstein distance FR
200	ssimulacra2	ssimulacra2	img/vid +ref	SSIMULACRA 2 (0-100, lower=better, JPEG XL standard)
201	butteraugli	butteraugli	img/vid +ref	Butteraugli perceptual distance (lower=better)
202	flip_score	flip	img/vid +ref	NVIDIA FLIP perceptual metric (0-1, lower=better)
203	vmaf_neg	vmaf_neg	img/vid +ref	VMAF NEG (no enhancement gain, 0-100, higher=better)
204	ilniqe	ilniqe	img/vid	IL-NIQE Integrated Local NIQE (lower=better)
205	nrqm	nrqm	img/vid	NRQM No-Reference Quality Metric (higher=better)
206	pi_score	pi	img/vid	Perceptual Index (PIRM challenge, lower=better)
207	piqe	piqe	img/vid	PIQE perception-based NR-IQA (lower=better)
208	maclip_score	maclip	img/vid	MACLIP multi-attribute CLIP NR-IQA (higher=better)
209	dmm	dmm	img/vid +ref	DMM Detail Model Metric FR (higher=better)
210	wadiqam_fr	wadiqam_fr	img/vid +ref	WaDIQaM full-reference (higher=better)
211	ssimc	ssimc	img/vid +ref	Complex Wavelet SSIM-C FR (higher=better)
212	cambi	cambi	img/vid	CAMBI banding index (0-24, lower=better)
213	xpsnr	xpsnr	img +ref	XPSNR perceptual PSNR (dB, higher=better)
214	vmaf_phone	vmaf_phone	img/vid +ref	VMAF phone model (0-100, higher=better)
215	vmaf_4k	vmaf_4k	img/vid +ref	VMAF 4K model (0-100, higher=better)
216	visqol	visqol	audio +ref	ViSQOL audio quality MOS (1-5, higher=better)
217	dnsmos_overall	dnsmos	audio	DNSMOS overall MOS (1-5, higher=better)
218	dnsmos_sig	dnsmos	audio	DNSMOS signal quality (1-5, higher=better)
219	dnsmos_bak	dnsmos	audio	DNSMOS background quality (1-5, higher=better)
220	pu_psnr	pu_metrics	img/vid +ref	PU-PSNR perceptually uniform HDR (dB, higher=better)
221	pu_ssim	pu_metrics	img/vid +ref	PU-SSIM perceptually uniform HDR (0-1, higher=better)
222	max_fall	hdr_metadata	img/vid	MaxFALL frame average light level (nits)
223	max_cll	hdr_metadata	img/vid	MaxCLL content light level (nits)
224	hdr_vdp	hdr_vdp	img/vid +ref	HDR-VDP visual difference predictor (higher=better)
225	delta_ictcp	delta_ictcp	img/vid +ref	Delta ICtCp HDR color difference (lower=better)
226	ciede2000	ciede2000	img/vid +ref	CIEDE2000 perceptual color difference (lower=better)
227	psnr_hvs	psnr_hvs	img/vid +ref	PSNR-HVS perceptually weighted (dB, higher=better)
228	psnr_hvs_m	psnr_hvs	img +ref	PSNR-HVS-M with masking (dB, higher=better)
229	cgvqm	cgvqm	img/vid +ref	CGVQM gaming quality (higher=better)
230	strred	strred	img +ref	STRRED reduced-reference temporal (lower=better)
231	p1203_mos	p1203	audio	ITU-T P.1203 streaming QoE MOS (1-5)
232	nemo_quality_score	nemo_curator	img +cap	Caption text quality (0-1)
233	nemo_quality_label	nemo_curator	img +cap	Quality label (Low/Medium/High)
234	human_fidelity_score	human_fidelity	img/vid	Body/hand/face quality (0-1, higher=better)
235	physics_score	physics	vid	Physics plausibility (0-1, higher=better)
236	commonsense_score	commonsense	img/vid	Common sense adherence (0-1, higher=better)
237	creativity_score	creativity	img/vid	Artistic novelty (0-1, higher=better)
238	chronomagic_mt_score	chronomagic	vid	Metamorphic temporal (0-1, higher=better)
239	chronomagic_ch_score	chronomagic	vid	Chrono-hallucination (0-1, lower=fewer)
240	compbench_attribute	t2v_compbench	vid +cap	Attribute binding (0-1)
241	compbench_object_rel	t2v_compbench	vid +cap	Object relationship (0-1)
242	compbench_action	t2v_compbench	vid +cap	Action binding (0-1)
243	compbench_spatial	t2v_compbench	vid +cap	Spatial relationship (0-1)
244	compbench_numeracy	t2v_compbench	vid +cap	Generative numeracy (0-1)
245	compbench_scene	t2v_compbench	vid +cap	Scene composition (0-1)
246	compbench_overall	t2v_compbench	vid +cap	Overall composition (0-1)

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.