cjones/immich
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'main' into rknn-toolkit-lite2

Browse Source
This commit is contained in:
Yoni Yang
2025-01-22 11:43:59 +08:00
committed by GitHub
parent dd52c2d68c c8abe9a2fd
commit 794da29411
130 changed files with 2154 additions and 2380 deletions
Download Patch File Download Diff File Expand all files Collapse all files
machine-learning/app/test_main.py
+28 -17
View File
@@ -10,6 +10,7 @@ from unittest import mock
import cv2
import numpy as np
import onnxruntime as ort
import orjson
import pytest
from fastapi import HTTPException
from fastapi.testclient import TestClient
@@ -396,11 +397,11 @@ class TestCLIP:
mocked.run.return_value = [[self.embedding]]
clip_encoder = OpenClipVisualEncoder("ViT-B-32__openai", cache_dir="test_cache")
embedding = clip_encoder.predict(pil_image)
assert isinstance(embedding, np.ndarray)
assert embedding.shape[0] == clip_model_cfg["embed_dim"]
assert embedding.dtype == np.float32
embedding_str = clip_encoder.predict(pil_image)
assert isinstance(embedding_str, str)
embedding = orjson.loads(embedding_str)
assert isinstance(embedding, list)
assert len(embedding) == clip_model_cfg["embed_dim"]
mocked.run.assert_called_once()
def test_basic_text(
@@ -418,11 +419,11 @@ class TestCLIP:
mocker.patch("app.models.clip.textual.Tokenizer.from_file", autospec=True)
clip_encoder = OpenClipTextualEncoder("ViT-B-32__openai", cache_dir="test_cache")
embedding = clip_encoder.predict("test search query")
assert isinstance(embedding, np.ndarray)
assert embedding.shape[0] == clip_model_cfg["embed_dim"]
assert embedding.dtype == np.float32
embedding_str = clip_encoder.predict("test search query")
assert isinstance(embedding_str, str)
embedding = orjson.loads(embedding_str)
assert isinstance(embedding, list)
assert len(embedding) == clip_model_cfg["embed_dim"]
mocked.run.assert_called_once()
def test_openclip_tokenizer(
@@ -558,8 +559,11 @@ class TestFaceRecognition:
assert isinstance(face.get("boundingBox"), dict)
assert set(face["boundingBox"]) == {"x1", "y1", "x2", "y2"}
assert all(isinstance(val, np.float32) for val in face["boundingBox"].values())
assert isinstance(face.get("embedding"), np.ndarray)
assert face["embedding"].shape[0] == 512
embedding_str = face.get("embedding")
assert isinstance(embedding_str, str)
embedding = orjson.loads(embedding_str)
assert isinstance(embedding, list)
assert len(embedding) == 512
assert isinstance(face.get("score", None), np.float32)
rec_model.get_feat.assert_called_once()
@@ -930,8 +934,10 @@ class TestPredictionEndpoints:
actual = response.json()
assert response.status_code == 200
assert isinstance(actual, dict)
assert isinstance(actual.get("clip", None), list)
assert np.allclose(expected, actual["clip"])
embedding = actual.get("clip", None)
assert isinstance(embedding, str)
parsed_embedding = orjson.loads(embedding)
assert np.allclose(expected, parsed_embedding)
def test_clip_text_endpoint(self, responses: dict[str, Any], deployed_app: TestClient) -> None:
expected = responses["clip"]["text"]
@@ -951,8 +957,10 @@ class TestPredictionEndpoints:
actual = response.json()
assert response.status_code == 200
assert isinstance(actual, dict)
assert isinstance(actual.get("clip", None), list)
assert np.allclose(expected, actual["clip"])
embedding = actual.get("clip", None)
assert isinstance(embedding, str)
parsed_embedding = orjson.loads(embedding)
assert np.allclose(expected, parsed_embedding)
def test_face_endpoint(self, pil_image: Image.Image, responses: dict[str, Any], deployed_app: TestClient) -> None:
byte_image = BytesIO()
@@ -983,5 +991,8 @@ class TestPredictionEndpoints:
for expected_face, actual_face in zip(responses["facial-recognition"], actual["facial-recognition"]):
assert expected_face["boundingBox"] == actual_face["boundingBox"]
assert np.allclose(expected_face["embedding"], actual_face["embedding"])
embedding = actual_face.get("embedding", None)
assert isinstance(embedding, str)
parsed_embedding = orjson.loads(embedding)
assert np.allclose(expected_face["embedding"], parsed_embedding)
assert np.allclose(expected_face["score"], actual_face["score"])