Merge pull request #1440 from serengil/feat-task-2002-post-batch-changes

post batch changes

deepface/DeepFace.py

@@ -166,7 +166,7 @@ def verify(
 
 
 def analyze(
-    img_path: Union[str, np.ndarray, IO[bytes]],
+    img_path: Union[str, np.ndarray, IO[bytes], List[str], List[np.ndarray], List[IO[bytes]]],
     actions: Union[tuple, list] = ("emotion", "age", "gender", "race"),
     enforce_detection: bool = True,
     detector_backend: str = "opencv",
@@ -178,7 +178,7 @@ def analyze(
     """
     Analyze facial attributes such as age, gender, emotion, and race in the provided image.
     Args:
-        img_path (str or np.ndarray or IO[bytes]): The exact path to the image, a numpy array
+        img_path (str, np.ndarray, IO[bytes], list): The exact path to the image, a numpy array
             in BGR format, a file object that supports at least `.read` and is opened in binary
             mode, or a base64 encoded image. If the source image contains multiple faces,
             the result will include information for each detected face.

deepface/models/FacialRecognition.py

@@ -11,6 +11,7 @@ else:
 
 # Notice that all facial recognition models must be inherited from this class
 
+
 # pylint: disable=too-few-public-methods
 class FacialRecognition(ABC):
     model: Union[Model, Any]
@@ -24,11 +25,24 @@ class FacialRecognition(ABC):
                 "You must overwrite forward method if it is not a keras model,"
                 f"but {self.model_name} not overwritten!"
             )
-        # model.predict causes memory issue when it is called in a for loop
-        # embedding = model.predict(img, verbose=0)[0].tolist()
-        if img.shape == 4 and img.shape[0] == 1:
-            img = img[0]
-        embeddings = self.model(img, training=False).numpy()
+
+        # predict expexts e.g. (1, 224, 224, 3) shaped inputs
+        if img.ndim == 3:
+            img = np.expand_dims(img, axis=0)
+
+        if img.ndim == 4 and img.shape[0] == 1:
+            # model.predict causes memory issue when it is called in a for loop
+            # embedding = model.predict(img, verbose=0)[0].tolist()
+            embeddings = self.model(img, training=False).numpy()
+        elif img.ndim == 4 and img.shape[0] > 1:
+            embeddings = self.model.predict_on_batch(img)
+        else:
+            raise ValueError(f"Input image must be (1, X, X, 3) shaped but it is {img.shape}")
+
+        assert isinstance(
+            embeddings, np.ndarray
+        ), f"Embeddings must be numpy array but it is {type(embeddings)}"
+
         if embeddings.shape[0] == 1:
             return embeddings[0].tolist()
         return embeddings.tolist()

deepface/models/facial_recognition/VGGFace.py

@@ -42,6 +42,7 @@ WEIGHTS_URL = (
     "https://github.com/serengil/deepface_models/releases/download/v1.0/vgg_face_weights.h5"
 )
 
+
 # pylint: disable=too-few-public-methods
 class VggFaceClient(FacialRecognition):
     """
@@ -70,10 +71,7 @@ class VggFaceClient(FacialRecognition):
         # having normalization layer in descriptor troubles for some gpu users (e.g. issue 957, 966)
         # instead we are now calculating it with traditional way not with keras backend
         embedding = super().forward(img)
-        if (
-            isinstance(embedding, list) and
-            isinstance(embedding[0], list)
-        ):
+        if isinstance(embedding, list) and len(embedding) > 0 and isinstance(embedding[0], list):
             embedding = verification.l2_normalize(embedding, axis=1)
         else:
             embedding = verification.l2_normalize(embedding)

deepface/modules/demography.py

@@ -1,5 +1,5 @@
 # built-in dependencies
-from typing import Any, Dict, List, Union
+from typing import Any, Dict, List, Union, IO
 
 # 3rd party dependencies
 import numpy as np
@@ -11,7 +11,7 @@ from deepface.models.demography import Gender, Race, Emotion
 
 
 def analyze(
-    img_path: Union[str, np.ndarray],
+    img_path: Union[str, np.ndarray, IO[bytes], List[str], List[np.ndarray], List[IO[bytes]]],
     actions: Union[tuple, list] = ("emotion", "age", "gender", "race"),
     enforce_detection: bool = True,
     detector_backend: str = "opencv",
@@ -19,14 +19,14 @@ def analyze(
     expand_percentage: int = 0,
     silent: bool = False,
     anti_spoofing: bool = False,
-) -> List[Dict[str, Any]]:
+) -> Union[List[Dict[str, Any]], List[List[Dict[str, Any]]]]:
     """
     Analyze facial attributes such as age, gender, emotion, and race in the provided image.
 
     Args:
-        img_path (str or np.ndarray): The exact path to the image, a numpy array in BGR format,
-            or a base64 encoded image. If the source image contains multiple faces, the result will
-            include information for each detected face.
+        img_path (str, np.ndarray, IO[bytes], list): The exact path to the image,
+            a numpy array in BGR format, or a base64 encoded image. If the source image
+            contains multiple faces, the result will include information for each detected face.
 
         actions (tuple): Attributes to analyze. The default is ('age', 'gender', 'emotion', 'race').
             You can exclude some of these attributes from the analysis if needed.
@@ -100,28 +100,28 @@ def analyze(
                - 'white': Confidence score for White ethnicity.
     """
 
-    if isinstance(img_path, np.ndarray) and len(img_path.shape) == 4:
-        # Received 4-D array, which means image batch.
-        # Check batch dimension and process each image separately.
-        if img_path.shape[0] > 1:
-            batch_resp_obj = []
-            # Execute analysis for each image in the batch.
-            for single_img in img_path:
-                # Call the analyze function for each image in the batch.
-                resp_obj = analyze(
-                    img_path=single_img,
-                    actions=actions,
-                    enforce_detection=enforce_detection,
-                    detector_backend=detector_backend,
-                    align=align,
-                    expand_percentage=expand_percentage,
-                    silent=silent,
-                    anti_spoofing=anti_spoofing,
-                )
+    # batch input
+    if (isinstance(img_path, np.ndarray) and img_path.ndim == 4 and img_path.shape[0] > 1) or (
+        isinstance(img_path, list)
+    ):
+        batch_resp_obj = []
+        # Execute analysis for each image in the batch.
+        for single_img in img_path:
+            # Call the analyze function for each image in the batch.
+            resp_obj = analyze(
+                img_path=single_img,
+                actions=actions,
+                enforce_detection=enforce_detection,
+                detector_backend=detector_backend,
+                align=align,
+                expand_percentage=expand_percentage,
+                silent=silent,
+                anti_spoofing=anti_spoofing,
+            )
 
-                # Append the response object to the batch response list.
-                batch_resp_obj.append(resp_obj)
-            return batch_resp_obj
+            # Append the response object to the batch response list.
+            batch_resp_obj.append(resp_obj)
+        return batch_resp_obj
 
     # if actions is passed as tuple with single item, interestingly it becomes str here
     if isinstance(actions, str):

deepface/modules/representation.py

@@ -1,5 +1,6 @@
 # built-in dependencies
 from typing import Any, Dict, List, Union, Optional, Sequence, IO
+from collections import defaultdict
 
 # 3rd party dependencies
 import numpy as np
@@ -161,17 +162,16 @@ def represent(
     # Forward pass through the model for the entire batch
     embeddings = model.forward(batch_images)
 
-    for idx in range(0, len(images)):
-        resp_obj = []
-        for idy, batch_index in enumerate(batch_indexes):
-            if idx == batch_index:
-                resp_obj.append(
-                    {
-                        "embedding": embeddings if len(batch_images) == 1 else embeddings[idy],
-                        "facial_area": batch_regions[idy],
-                        "face_confidence": batch_confidences[idy],
-                    }
-                )
-        resp_objs.append(resp_obj)
+    resp_objs_dict = defaultdict(list)
+    for idy, batch_index in enumerate(batch_indexes):
+        resp_objs_dict[batch_index].append(
+            {
+                "embedding": embeddings if len(batch_images) == 1 else embeddings[idy],
+                "facial_area": batch_regions[idy],
+                "face_confidence": batch_confidences[idy],
+            }
+        )
+
+    resp_objs = [resp_objs_dict[idx] for idx in range(len(images))]
 
     return resp_objs[0] if len(images) == 1 else resp_objs

tests/test_analyze.py

@@ -16,9 +16,13 @@ detectors = ["opencv", "mtcnn"]
 def test_standard_analyze():
     img = "dataset/img4.jpg"
     demography_objs = DeepFace.analyze(img, silent=True)
+
+    # return type should be list of dict for non batch input
+    assert isinstance(demography_objs, list)
+
     for demography in demography_objs:
+        assert isinstance(demography, dict)
         logger.debug(demography)
-        assert type(demography) == dict
         assert demography["age"] > 20 and demography["age"] < 40
         assert demography["dominant_gender"] == "Woman"
     logger.info("✅ test standard analyze done")
@@ -99,9 +103,13 @@ def test_analyze_for_some_actions():
 def test_analyze_for_preloaded_image():
     img = cv2.imread("dataset/img1.jpg")
     resp_objs = DeepFace.analyze(img, silent=True)
+
+    # return type should be list of dict for non batch input
+    assert isinstance(resp_objs, list)
+
     for resp_obj in resp_objs:
+        assert isinstance(resp_obj, dict)
         logger.debug(resp_obj)
-        assert type(resp_obj) == dict
         assert resp_obj["age"] > 20 and resp_obj["age"] < 40
         assert resp_obj["dominant_gender"] == "Woman"
 
@@ -127,7 +135,10 @@ def test_analyze_for_different_detectors():
             results = DeepFace.analyze(
                 img_path, actions=("gender",), detector_backend=detector, enforce_detection=False
             )
+            # return type should be list of dict for non batch input
+            assert isinstance(results, list)
             for result in results:
+                assert isinstance(result, dict)
                 logger.debug(result)
 
                 # validate keys
@@ -138,13 +149,63 @@ def test_analyze_for_different_detectors():
                 ]
 
                 # validate probabilities
-                assert type(result) == dict
                 if result["dominant_gender"] == "Man":
                     assert result["gender"]["Man"] > result["gender"]["Woman"]
                 else:
                     assert result["gender"]["Man"] < result["gender"]["Woman"]
 
 
+def test_analyze_for_batched_image_as_list_of_string():
+    img_paths = ["dataset/img1.jpg", "dataset/img2.jpg", "dataset/couple.jpg"]
+    expected_faces = [1, 1, 2]
+
+    demography_batch = DeepFace.analyze(img_path=img_paths, silent=True)
+    # return type should be list of list of dict for batch input
+    assert isinstance(demography_batch, list)
+
+    # 3 image in batch, so 3 demography objects
+    assert len(demography_batch) == len(img_paths)
+
+    for idx, demography_objs in enumerate(demography_batch):
+        assert isinstance(demography_objs, list)
+        assert len(demography_objs) == expected_faces[idx]
+        for demography_obj in demography_objs:
+            assert isinstance(demography_obj, dict)
+
+            assert demography_obj["age"] > 20 and demography_obj["age"] < 40
+            assert demography_obj["dominant_gender"] in ["Woman", "Man"]
+
+    logger.info("✅ test analyze for batched image as list of string done")
+
+
+def test_analyze_for_batched_image_as_list_of_numpy():
+    img_paths = ["dataset/img1.jpg", "dataset/img2.jpg", "dataset/couple.jpg"]
+    expected_faces = [1, 1, 2]
+
+    imgs = []
+    for img_path in img_paths:
+        img = cv2.imread(img_path)
+        imgs.append(img)
+
+    demography_batch = DeepFace.analyze(img_path=imgs, silent=True)
+    # return type should be list of list of dict for batch input
+    assert isinstance(demography_batch, list)
+
+    # 3 image in batch, so 3 demography objects
+    assert len(demography_batch) == len(img_paths)
+
+    for idx, demography_objs in enumerate(demography_batch):
+        assert isinstance(demography_objs, list)
+        assert len(demography_objs) == expected_faces[idx]
+        for demography_obj in demography_objs:
+            assert isinstance(demography_obj, dict)
+
+            assert demography_obj["age"] > 20 and demography_obj["age"] < 40
+            assert demography_obj["dominant_gender"] in ["Woman", "Man"]
+
+    logger.info("✅ test analyze for batched image as list of numpy done")
+
+
 def test_analyze_for_numpy_batched_image():
     img1_path = "dataset/img4.jpg"
     img2_path = "dataset/couple.jpg"
@@ -163,14 +224,20 @@ def test_analyze_for_numpy_batched_image():
     assert img.shape[0] == 2  # Check batch size.
 
     demography_batch = DeepFace.analyze(img, silent=True)
+    # return type should be list of list of dict for batch input
+
+    assert isinstance(demography_batch, list)
+
     # 2 image in batch, so 2 demography objects.
     assert len(demography_batch) == 2
 
     for i, demography_objs in enumerate(demography_batch):
+        assert isinstance(demography_objs, list)
 
         assert len(demography_objs) == expected_num_faces[i]
         for demography in demography_objs:  # Iterate over faces
-            assert isinstance(demography, dict)  # Check type
+            assert isinstance(demography, dict)
+
             assert demography["age"] > 20 and demography["age"] < 40
             assert demography["dominant_gender"] in ["Woman", "Man"]
 

tests/test_represent.py

@@ -183,6 +183,8 @@ def test_batched_represent_for_list_input(model_name):
         assert len(single_embedding_objs) == len(batched_embedding_objs[idx])
 
         for alpha, beta in zip(single_embedding_objs, batched_embedding_objs[idx]):
+            assert isinstance(alpha, dict)
+            assert isinstance(beta, dict)
             assert np.allclose(
                 alpha["embedding"], beta["embedding"], rtol=1e-2, atol=1e-2
             ), "Embeddings do not match within tolerance"