helpy4/immich
1
0
Fork 0
mirror of https://github.com/immich-app/immich synced 2025-11-07 17:27:20 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

feat(ml): composable ml (#9973)

Browse source 
* modularize model classes

* various fixes

* expose port

* change response

* round coordinates

* simplify preload

* update server

* simplify interface

simplify

* update tests

* composable endpoint

* cleanup

fixes

remove unnecessary interface

support text input, cleanup

* ew camelcase

* update server

server fixes

fix typing

* ml fixes

update locustfile

fixes

* cleaner response

* better repo response

* update tests

formatting and typing

rename

* undo compose change

* linting

fix type

actually fix typing

* stricter typing

fix detection-only response

no need for defaultdict

* update spec file

update api

linting

* update e2e

* unnecessary dimension

* remove commented code

* remove duplicate code

* remove unused imports

* add batch dim
This commit is contained in:
Mert 2024-06-06 23:09:47 -04:00 • committed by GitHub
parent 7a46f80ddc
commit 2b1b43a7e4
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
39 changed files with 982 additions and 999 deletions
Download patch file Download diff file Expand all files Collapse all files

72
machine-learning/app/schemas.py
View file

@ -1,5 +1,5 @@
from enum import Enum
from typing import Any, Protocol, TypedDict, TypeGuard
from typing import Any, Literal, Protocol, TypedDict, TypeGuard, TypeVar
import numpy as np
import numpy.typing as npt
@ -28,31 +28,87 @@ class BoundingBox(TypedDict):
y2: int
class ModelType(StrEnum):
CLIP = "clip"
class ModelTask(StrEnum):
FACIAL_RECOGNITION = "facial-recognition"
SEARCH = "clip"
class ModelRuntime(StrEnum):
ONNX = "onnx"
class ModelType(StrEnum):
DETECTION = "detection"
RECOGNITION = "recognition"
TEXTUAL = "textual"
VISUAL = "visual"
class ModelFormat(StrEnum):
ARMNN = "armnn"
ONNX = "onnx"
class ModelSource(StrEnum):
INSIGHTFACE = "insightface"
MCLIP = "mclip"
OPENCLIP = "openclip"
ModelIdentity = tuple[ModelType, ModelTask]
class ModelSession(Protocol):
def run(
self,
output_names: list[str] | None,
input_feed: dict[str, npt.NDArray[np.float32]] | dict[str, npt.NDArray[np.int32]],
run_options: Any = None,
) -> list[npt.NDArray[np.float32]]: ...
class HasProfiling(Protocol):
profiling: dict[str, float]
class Face(TypedDict):
class FaceDetectionOutput(TypedDict):
boxes: npt.NDArray[np.float32]
scores: npt.NDArray[np.float32]
landmarks: npt.NDArray[np.float32]
class DetectedFace(TypedDict):
boundingBox: BoundingBox
embedding: npt.NDArray[np.float32]
imageWidth: int
imageHeight: int
score: float
FacialRecognitionOutput = list[DetectedFace]
class PipelineEntry(TypedDict):
modelName: str
options: dict[str, Any]
PipelineRequest = dict[ModelTask, dict[ModelType, PipelineEntry]]
class InferenceEntry(TypedDict):
name: str
task: ModelTask
type: ModelType
options: dict[str, Any]
InferenceEntries = tuple[list[InferenceEntry], list[InferenceEntry]]
InferenceResponse = dict[ModelTask | Literal["imageHeight"] | Literal["imageWidth"], Any]
def has_profiling(obj: Any) -> TypeGuard[HasProfiling]:
return hasattr(obj, "profiling") and isinstance(obj.profiling, dict)
def is_ndarray(obj: Any, dtype: "type[np._DTypeScalar_co]") -> "TypeGuard[npt.NDArray[np._DTypeScalar_co]]":
return isinstance(obj, np.ndarray) and obj.dtype == dtype
T = TypeVar("T")