immich/machine-learning/immich_ml/models/__init__.py

49 lines
2 KiB
Python
Raw Normal View History

from typing import Any
from immich_ml.models.base import InferenceModel
from immich_ml.models.clip.textual import MClipTextualEncoder, OpenClipTextualEncoder
from immich_ml.models.clip.visual import OpenClipVisualEncoder
feat: ocr (#18836) * feat: add OCR functionality and related configurations * chore: update labeler configuration for machine learning files * feat(i18n): enhance OCR model descriptions and add orientation classification and unwarping features * chore: update Dockerfile to include ccache for improved build performance * feat(ocr): enhance OCR model configuration with orientation classification and unwarping options, update PaddleOCR integration, and improve response structure * refactor(ocr): remove OCR_CLEANUP job from enum and type definitions * refactor(ocr): remove obsolete OCR entity and migration files, and update asset job status and schema to accommodate new OCR table structure * refactor(ocr): update OCR schema and response structure to use individual coordinates instead of bounding box, and adjust related service and repository files * feat: enhance OCR configuration and functionality - Updated OCR settings to include minimum detection box score, minimum detection score, and minimum recognition score. - Refactored PaddleOCRecognizer to utilize new scoring parameters. - Introduced new database tables for asset OCR data and search functionality. - Modified related services and repositories to support the new OCR features. - Updated translations for improved clarity in settings UI. * sql changes * use rapidocr * change dto * update web * update lock * update api * store positions as normalized floats * match column order in db * update admin ui settings descriptions fix max resolution key set min threshold to 0.1 fix bind * apply config correctly, adjust defaults * unnecessary model type * unnecessary sources * fix(ocr): switch RapidOCR lang type from LangDet to LangRec * fix(ocr): expose lang_type (LangRec.CH) and font_path on OcrOptions for RapidOCR * fix(ocr): make OCR text search case- and accent-insensitive using ILIKE + unaccent * fix(ocr): add OCR search fields * fix: Add OCR database migration and update ML prediction logic. * trigrams are already case insensitive * add tests * format * update migrations * wrong uuid function * linting * maybe fix medium tests * formatting * fix weblate check * openapi * sql * minor fixes * maybe fix medium tests part 2 * passing medium tests * format web * readd sql * format dart * disabled in e2e * chore: translation ordering --------- Co-authored-by: mertalev <101130780+mertalev@users.noreply.github.com> Co-authored-by: Alex Tran <alex.tran1502@gmail.com>
2025-10-27 22:09:55 +08:00
from immich_ml.models.ocr.detection import TextDetector
from immich_ml.models.ocr.recognition import TextRecognizer
from immich_ml.schemas import ModelSource, ModelTask, ModelType
from .constants import get_model_source
from .facial_recognition.detection import FaceDetector
from .facial_recognition.recognition import FaceRecognizer
def get_model_class(model_name: str, model_type: ModelType, model_task: ModelTask) -> type[InferenceModel]:
source = get_model_source(model_name)
match source, model_type, model_task:
case ModelSource.OPENCLIP | ModelSource.MCLIP, ModelType.VISUAL, ModelTask.SEARCH:
return OpenClipVisualEncoder
case ModelSource.OPENCLIP, ModelType.TEXTUAL, ModelTask.SEARCH:
return OpenClipTextualEncoder
case ModelSource.MCLIP, ModelType.TEXTUAL, ModelTask.SEARCH:
return MClipTextualEncoder
case ModelSource.INSIGHTFACE, ModelType.DETECTION, ModelTask.FACIAL_RECOGNITION:
return FaceDetector
case ModelSource.INSIGHTFACE, ModelType.RECOGNITION, ModelTask.FACIAL_RECOGNITION:
return FaceRecognizer
feat: ocr (#18836) * feat: add OCR functionality and related configurations * chore: update labeler configuration for machine learning files * feat(i18n): enhance OCR model descriptions and add orientation classification and unwarping features * chore: update Dockerfile to include ccache for improved build performance * feat(ocr): enhance OCR model configuration with orientation classification and unwarping options, update PaddleOCR integration, and improve response structure * refactor(ocr): remove OCR_CLEANUP job from enum and type definitions * refactor(ocr): remove obsolete OCR entity and migration files, and update asset job status and schema to accommodate new OCR table structure * refactor(ocr): update OCR schema and response structure to use individual coordinates instead of bounding box, and adjust related service and repository files * feat: enhance OCR configuration and functionality - Updated OCR settings to include minimum detection box score, minimum detection score, and minimum recognition score. - Refactored PaddleOCRecognizer to utilize new scoring parameters. - Introduced new database tables for asset OCR data and search functionality. - Modified related services and repositories to support the new OCR features. - Updated translations for improved clarity in settings UI. * sql changes * use rapidocr * change dto * update web * update lock * update api * store positions as normalized floats * match column order in db * update admin ui settings descriptions fix max resolution key set min threshold to 0.1 fix bind * apply config correctly, adjust defaults * unnecessary model type * unnecessary sources * fix(ocr): switch RapidOCR lang type from LangDet to LangRec * fix(ocr): expose lang_type (LangRec.CH) and font_path on OcrOptions for RapidOCR * fix(ocr): make OCR text search case- and accent-insensitive using ILIKE + unaccent * fix(ocr): add OCR search fields * fix: Add OCR database migration and update ML prediction logic. * trigrams are already case insensitive * add tests * format * update migrations * wrong uuid function * linting * maybe fix medium tests * formatting * fix weblate check * openapi * sql * minor fixes * maybe fix medium tests part 2 * passing medium tests * format web * readd sql * format dart * disabled in e2e * chore: translation ordering --------- Co-authored-by: mertalev <101130780+mertalev@users.noreply.github.com> Co-authored-by: Alex Tran <alex.tran1502@gmail.com>
2025-10-27 22:09:55 +08:00
case ModelSource.PADDLE, ModelType.DETECTION, ModelTask.OCR:
return TextDetector
case ModelSource.PADDLE, ModelType.RECOGNITION, ModelTask.OCR:
return TextRecognizer
case _:
raise ValueError(f"Unknown model combination: {source}, {model_type}, {model_task}")
def from_model_type(model_name: str, model_type: ModelType, model_task: ModelTask, **kwargs: Any) -> InferenceModel:
return get_model_class(model_name, model_type, model_task)(model_name, **kwargs)
def get_model_deps(model_name: str, model_type: ModelType, model_task: ModelTask) -> list[tuple[ModelType, ModelTask]]:
return get_model_class(model_name, model_type, model_task).depends