implement face detection; fix image reading

gigachadify.py

@@ -3,10 +3,12 @@ from pathlib import Path
 from PIL import Image, ImageOps
 import requests
 from io import BytesIO
+import matplotlib.pyplot as plt
 
 import numpy as np
 import torch
 import insightface
+import cv2
 
 
 GIGACHAD_IMAGE_PATH = "images/gigachad.png"
@@ -23,29 +25,35 @@ if torch.cuda is not None:
         PROVIDER = ["CPUExecutionProvider"]
 
 
-def get_face_analysis_model() -> torch.nn.Module:
+def get_face_analysis_model() -> insightface.app.FaceAnalysis:
     return insightface.app.FaceAnalysis(
-        name="buffalo_l", root=INSIGHTFACE_PATH, providers=PROVIDER
+        name="buffalo_l", providers=PROVIDER, root=INSIGHTFACE_PATH
     )
 
 
-def detect_faces(image: torch.Tensor) -> torch.Tensor:
-    raise NotImplementedError
+def detect_faces(
+    img_data: np.ndarray,
+    face_analyser: insightface.app.FaceAnalysis,
+    det_size=(640, 640),
+) -> list:
+    face_analyser.prepare(ctx_id=0, det_size=det_size)
+    return face_analyser.get(img_data)
 
 
 def face_detection_asserts(base_faces, target_faces):
     raise NotImplementedError
 
 
-def get_face_single(faces: torch.Tensor) -> torch.Tensor:
-    raise NotImplementedError
+def get_face_single(faces: torch.tensor) -> torch.tensor:
+    # TODO
+    return faces[0]
 
 
 def swap_faces(
-    base_face_img: torch.Tensor,
-    target_face_img: torch.Tensor,
-    face_analysis_model: torch.nn.Module,
-) -> torch.Tensor:
+    base_face_img: np.ndarray,
+    target_face_img: np.ndarray,
+    face_analysis_model: insightface.app.FaceAnalysis,
+) -> torch.tensor:
     base_faces = detect_faces(base_face_img, face_analysis_model)
     target_faces = detect_faces(target_face_img, face_analysis_model)
 
@@ -65,7 +73,7 @@ def is_url(image_path: str) -> bool:
     return image_path.startswith("http://") or image_path.startswith("https://")
 
 
-def read_image(image_path: str) -> torch.tensor:
+def read_image(image_path: str) -> np.ndarray:
     if is_url(image_path):
         response = requests.get(image_path)
         image = Image.open(BytesIO(response.content))
@@ -75,7 +83,8 @@ def read_image(image_path: str) -> torch.tensor:
     image = ImageOps.exif_transpose(image)
     image = image.convert("RGB")
     image = np.array(image).astype(np.float32) / 255.0
-    image = torch.from_numpy(image).unsqueeze(0)
+    image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
+    image = (image * 255).astype(np.uint8)
     return image