Merge pull request #15 from DocsaidLab/feat/add_face_attributes

Add face attributes and fix jsonable by update capybara-docsaid to 0.12.0

Author: kunkunlin1221

.github/workflows/release.yml

@@ -36,7 +36,7 @@ jobs:
       - name: Bump package version
         if: ${{ matrix.os == 'ubuntu-latest' || matrix.os == 'ubuntu-24.04-arm' }}
         run: |
-          sed -i "s/__version__ = '[0-9]\+\(\.[0-9]\+\)\{1,2\}\(rc[0-9]\+\|[ab][0-9]\+\)\?'/__version__ = '${{ github.event.inputs.version_tag }}'/g" pyface/__init__.py
+          sed -E -i "s/__version__[[:space:]]*=[[:space:]]*['\"][^'\"]+['\"]/__version__ = '${{ github.event.inputs.version_tag }}'/g" pyface/__init__.py
 
       - name: Bump package version
         if: ${{ matrix.os == 'macos-13' || matrix.os == 'macos-latest' }}

pyface/face_service.py

@@ -94,7 +94,6 @@ def _fill_results_to_faces_list(
             for face in faces:
                 if face.attribute is None:
                     face.attribute = Attribute()
-
                 if gender_results is not None:
                     face.attribute.gender = gender_results[i]["gender"]
                 if lmk_results is not None:
@@ -116,7 +115,7 @@ def _fill_results_to_faces_list(
                     dep_result = dep_results[i]
                     face.tddfa = TDDFA(
                         param=dep_result["param"],
-                        lmk68pt=dep_result["lmk3d68pt"],
+                        lmk3d68pt=dep_result["lmk3d68pt"],
                         depth_img=dep_result["depth_img"],
                         yaw=dep_result["pose_degree"][0],
                         roll=dep_result["pose_degree"][1],

pyface/object.py

@@ -4,7 +4,6 @@
 import capybara as cb
 import cv2
 import numpy as np
-from pybase64 import b64encode
 
 from .components.enums import FacePose, FakeType
 
@@ -30,19 +29,41 @@ class Eye(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
     is_open: Optional[bool] = field(default=None)
     score: Optional[float] = field(default=None)
 
+    @classmethod
+    def from_json(cls, data) -> "Eye":
+        return cls(
+            is_open=data.get("is_open"),
+            score=data.get("score"),
+        )
+
 
 @dataclass()
 class Mouth(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
     is_open: Optional[bool] = field(default=None)
     score: Optional[float] = field(default=None)
 
+    @classmethod
+    def from_json(cls, data) -> "Mouth":
+        return cls(
+            is_open=data.get("is_open"),
+            score=data.get("score"),
+        )
+
 
 @dataclass()
 class WhetherOrNot(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
     is_true: Optional[bool] = field(default=None)
     value: Optional[float] = field(default=None)
     threshold: Optional[float] = field(default=None)
 
+    @classmethod
+    def from_json(cls, data) -> "WhetherOrNot":
+        return cls(
+            is_true=data.get("is_true"),
+            value=data.get("value"),
+            threshold=data.get("threshold"),
+        )
+
 
 @dataclass()
 class Liveness(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
@@ -51,29 +72,64 @@ class Liveness(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
     threshold: Optional[Union[float, np.number]] = field(default=None)
     fake_type: Optional[FakeType] = field(default=None)
 
+    @classmethod
+    def from_json(cls, data):
+        return cls(
+            is_true=data.get("is_true"),
+            value=data.get("value"),
+            threshold=data.get("threshold"),
+            fake_type=FakeType(data["fake_type"]) if data.get("fake_type") is not None else None,
+        )
+
 
 @dataclass()
 class TDDFA(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
     param: Optional[np.ndarray] = field(default=None)
-    lmk68pt: Optional[np.ndarray] = field(default=None)
+    lmk3d68pt: Optional[np.ndarray] = field(default=None)
     depth_img: Optional[np.ndarray] = field(default=None)
     yaw: Optional[float] = field(default=None)
     roll: Optional[float] = field(default=None)
     pitch: Optional[float] = field(default=None)
 
+    @classmethod
+    def from_json(cls, data) -> "TDDFA":
+        return cls(
+            param=cb.b64str_to_npy(data["param"]) if data.get("param") is not None else None,
+            lmk3d68pt=np.array(data["lmk3d68pt"]) if data.get("lmk3d68pt") is not None else None,
+            depth_img=cb.b64str_to_img(data["depth_img"]) if data.get("depth_img") is not None else None,
+            yaw=data.get("yaw"),
+            roll=data.get("roll"),
+            pitch=data.get("pitch"),
+        )
+
 
 @dataclass()
 class Encode(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
     vector: Optional[np.ndarray] = field(default=None)
     version: Optional[str] = field(default=None)
 
+    @classmethod
+    def from_json(cls, data) -> "Encode":
+        return cls(
+            vector=cb.b64str_to_npy(data["vector"]) if data.get("vector") is not None else None,
+            version=data.get("version"),
+        )
+
 
 @dataclass()
 class Who(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
     name: Optional[str] = field(default="?")
     confidence: Optional[float] = field(default=None)
     recognized_level: Optional[int] = field(default=None)
 
+    @classmethod
+    def from_json(cls, data) -> "Who":
+        return cls(
+            name=data.get("name", "?"),
+            confidence=data.get("confidence"),
+            recognized_level=data.get("recognized_level"),
+        )
+
 
 @dataclass()
 class Attribute(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
@@ -85,6 +141,18 @@ class Attribute(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
     right_eye: Optional[Eye] = field(default=None)
     mouth: Optional[Mouth] = field(default=None)
 
+    @classmethod
+    def from_json(cls, data) -> "Attribute":
+        return cls(
+            age=data.get("age"),
+            gender=data.get("gender"),
+            race=data.get("race"),
+            pose=FacePose.obj_to_enum(data["pose"]) if data.get("pose") is not None else None,
+            left_eye=Eye.from_json(data["left_eye"]) if data.get("left_eye") is not None else None,
+            right_eye=Eye.from_json(data["right_eye"]) if data.get("right_eye") is not None else None,
+            mouth=Mouth.from_json(data["mouth"]) if data.get("mouth") is not None else None,
+        )
+
 
 @dataclass()
 class Face(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
@@ -98,16 +166,34 @@ class Face(cb.DataclassToJsonMixin, cb.DataclassCopyMixin):
     lmk106pt: Optional[cb.Keypoints] = field(default=None)
     liveness: Optional[Liveness] = field(default=None)
     attribute: Optional[Attribute] = field(default=None)
+    # assign jsonable functions for some fields
     jsonable_func = {
-        "vector": lambda x: b64encode(x.astype("float32").tobytes()).decode("utf-8") if x is not None else None,
+        "vector": lambda x: cb.npy_to_b64str(x) if x is not None else None,
         "norm_img": lambda x: cb.img_to_b64str(x, cb.IMGTYP.PNG) if x is not None else None,
+        "depth_img": lambda x: cb.img_to_b64str(x, cb.IMGTYP.PNG) if x is not None else None,
+        "param": lambda x: cb.npy_to_b64str(x) if x is not None else None,
     }
     # pose: Optional[FacePose] = field(default=None)
     # blur: Optional[WhetherOrNot] = field(default=None)
     # occlusion: Optional[Occlusion] = field(default=None)
-    # lmk68pt: Optional[cb.Keypoints] = field(default=None)
+    # lmk3d68pt: Optional[cb.Keypoints] = field(default=None)
     # analysis_infos: Optional[dict] = field(default=None)
 
+    @classmethod
+    def from_json(cls, data: dict) -> "Face":
+        return cls(
+            box=cb.Box(data["box"]),
+            score=data["score"],
+            lmk5pt=cb.Keypoints(np.array(data["lmk5pt"])) if data.get("lmk5pt") is not None else None,
+            norm_img=cb.b64str_to_img(data["norm_img"]) if data.get("norm_img") is not None else None,
+            tddfa=TDDFA.from_json(data["tddfa"]) if data.get("tddfa") is not None else None,
+            encoding=Encode.from_json(data["encoding"]) if data.get("encoding") is not None else None,
+            who=Who.from_json(data["who"]) if data.get("who") is not None else None,
+            lmk106pt=cb.Keypoints(np.array(data["lmk106pt"])) if data.get("lmk106pt") is not None else None,
+            liveness=Liveness.from_json(data["liveness"]) if data.get("liveness") is not None else None,
+            attribute=Attribute.from_json(data["attribute"]) if data.get("attribute") is not None else None,
+        )
+
 
 ATTR_NAMES = [f.name for f in fields(Face)]
 
@@ -252,12 +338,19 @@ def gen_info_img(self, mosaic_face: bool = False):
         return img
 
     def be_jsonable(self):
-        raw_image = cb.img_to_b64(self.raw_image, cb.IMGTYP.PNG).decode("utf-8") if self.raw_image is not None else None
+        raw_image = cb.img_to_b64str(self.raw_image, cb.IMGTYP.PNG) if self.raw_image is not None else None
         return {
             "raw_image": raw_image,
             "faces": [x.be_jsonable() for x in self.faces],
         }
 
+    @classmethod
+    def from_json(cls, data: dict) -> "Faces":
+        return cls(
+            raw_image=cb.b64str_to_img(data["raw_image"]) if data.get("raw_image") is not None else None,
+            faces=[Face.from_json(x) for x in data.get("faces", [])],
+        )
+
 
 # def _remove_none_in_jsonized_face(jsonized_face: dict) -> dict:
 #     outs = {}

pyproject.toml

@@ -31,7 +31,7 @@ classifiers = [
 dependencies = [
   "numpy>=2",                 # 執行期也支援 NumPy 2
   "scikit-image",
-  "capybara-docsaid",
+  "capybara-docsaid>=0.12.0",
   "huggingface_hub"
 ]
 

tests/resources/answer/EmmaWatson1_jsonable.json

@@ -0,0 +1,46 @@
+import capybara as cb
+import pytest
+
+import pyface as pf
+
+RESOURCE_DIR = cb.get_curdir(__file__) / "resources"
+ANSWER_DIR = RESOURCE_DIR / "answer"
+ANSWER_DIR.mkdir(exist_ok=True, parents=True)
+
+TEST_DATA = [
+    {
+        "img_fpath": RESOURCE_DIR / "EmmaWatson1.jpg",
+        "json_fpath": ANSWER_DIR / "EmmaWatson1_jsonable.json",
+    },
+    {
+        "img_fpath": RESOURCE_DIR / "JohnnyDepp1.jpg",
+        "json_fpath": ANSWER_DIR / "JohnnyDepp1_jsonable.json",
+    },
+]
+
+
+@pytest.mark.parametrize("data", TEST_DATA)
+def test_jsonable(data):
+    expected = cb.load_json(data["json_fpath"])
+    faces = pf.Faces.from_json(expected)
+    output = faces.be_jsonable()
+    assert output == expected
+
+
+def gen_data(data):
+    face_service = pf.FaceService(
+        enable_depth=True,
+        enable_landmark=True,
+        enable_recognition=True,
+        enable_gender=True,
+        face_bank=RESOURCE_DIR / "face_bank",
+    )
+    img = cb.imread(data["img_fpath"])
+    faces = face_service([img], do_1n=True)[0]
+    output = faces.be_jsonable()
+    cb.dump_json(output, data["expected"])
+
+
+if __name__ == "__main__":
+    for data in TEST_DATA:
+        gen_data(data)