fix

demo.py

@@ -1,16 +1,45 @@
 
 from image_input_handler import  UniversalImageInputHandler
 # from facexformer_pipeline import facexformer_pipeline
-
-from facexformer_pipeline.facexformer_pipeline import facexformer_pipeline
+from facexformer_pipeline.facexformer_pipeline import FacexformerPipeline
+from visual_debugger import VisualDebugger, Annotation, AnnotationType
+import numpy as np
 
 def main():
-
+    vd = VisualDebugger(tag="facexformer", debug_folder_path="./", active=True)
     image_path = "sample_image.jpg"
     uih = UniversalImageInputHandler(image_path, debug=False)
     COMPATIBLE, img = uih.COMPATIBLE, uih.img
+    print('COMPATIBLE:', COMPATIBLE)
+
+    pipeline = FacexformerPipeline(tasks=['landmark', 'headpose'], debug=True)
+    results = pipeline.run_model(uih.img)
+
+    transformed_image = results['transformed_image'].numpy()
+    transformed_image = np.transpose(transformed_image, (1, 2, 0))
+    unnormalized_image = results['image']
+
+    # print(results['headpose'])
+
+    landmarks_annotation = [Annotation(type=AnnotationType.POINTS,  coordinates=results["landmark_list"], color=(0, 255, 0))]
+    scaled_landmarks_annotation = [ Annotation(type=AnnotationType.POINTS, coordinates=results["scaled_landmarks"], color=(0, 255, 0))]
+
+    vd.visual_debug(img, landmarks_annotation, process_step="landmarks" )
+    vd.visual_debug(img, scaled_landmarks_annotation, process_step="scaled_landmarks")
+
+    vd.visual_debug(transformed_image, landmarks_annotation, process_step="landmarks_on_transformed")
+    vd.visual_debug(unnormalized_image, landmarks_annotation, process_step="landmarks_on_unnormalized_image ")
+
+    vd.visual_debug(transformed_image, scaled_landmarks_annotation, process_step="landmarks_on_transformed")
+    vd.visual_debug(unnormalized_image, scaled_landmarks_annotation, process_step="landmarks_on_unnormalized_image ")
+
 
-    facexformer_pipeline(img)
+# annotations = [Annotation(type=AnnotationType.PITCH_YAW_ROLL,
+    #                           coordinates=None,
+    #                           orientation=(
+    #                           results["headpose"]["pitch"], results["headpose"]["yaw"], results["headpose"]["roll"]),
+    #                           color=(0, 255, 0))]
+    # vd.visual_debug(img, annotations, process_step="head_orientation")
 
 if __name__ == "__main__":
 

facexformer_pipeline/facexformer_pipeline.py

@@ -11,7 +11,7 @@
 from network import FaceXFormer
 from facenet_pytorch import MTCNN
 from image_input_handler import  UniversalImageInputHandler
-from utils import denorm_points, unnormalize, adjust_bbox, visualize_head_pose, visualize_landmarks, visualize_mask
+from facexformer_pipeline.utils import denorm_points, unnormalize, adjust_bbox, visualize_head_pose, visualize_landmarks, visualize_mask
 from task_postprocesser import task_faceparsing,  process_landmarks, task_headpose , task_attributes, task_gender, process_visibility
 import torchvision.transforms as transforms
 from huggingface_hub import hf_hub_download
@@ -124,9 +124,32 @@ def process_task_output(self, task_id, output, results, images):
             results['age_gender_race_dict'] = task_gender(output[4], output[5], output[6])
         elif task_id == 5:
             results['visibility'] = process_visibility(output[3])
+
+    # from PIL import Image, ImageDraw
+    # import numpy as np
+
+    def scale_landmarks_to_original_image(self, original_image, landmarks, resized_image_size=(224, 224)):
+        # print("landmarks[0]:", landmarks[0])
+        original_width, original_height = original_image.shape[1],  original_image.shape[0]
+        # print("original_width:", original_width, "original_height:", original_height)
+        resized_width, resized_height = resized_image_size
+        scale_x = original_width / resized_width
+        scale_y = original_height / resized_height
+
+        # Scale landmarks back to original image dimensions
+
+
+        scaled_landmarks = [(x * scale_x, y * scale_y) for (x, y) in landmarks]
+        scaled_landmarks_int = [(int(round(x)), int(round(y))) for (x, y) in scaled_landmarks]
+
+        # print("scaled_landmarks[0]:", scaled_landmarks[0])
+
+        return scaled_landmarks_int
+
+
     def run_model(self, image, image_is_cropped=True):
         results = {}
-
+        original_image=image.copy()
         image = Image.fromarray(image)
         if not image_is_cropped:
             image = self.crop_face_area_from_image(image)
@@ -143,6 +166,9 @@ def run_model(self, image, image_is_cropped=True):
         image = (image * 255).astype(np.uint8)
         image = image[:, :, ::-1]
         results['image'] = image
+        results['transformed_image'] = model_ready_image[0]
+
+        results['scaled_landmarks'] =self.scale_landmarks_to_original_image(original_image,results['landmark_list'] )
         return results
 
 def main():

facexformer_pipeline/task_postprocesser.py

@@ -12,7 +12,7 @@
 from facenet_pytorch import MTCNN
 import os
 # from argparse import  args
-from utils import denorm_points, unnormalize, adjust_bbox, visualize_head_pose, visualize_landmarks, visualize_mask
+from facexformer_pipeline.utils import denorm_points, unnormalize, adjust_bbox, visualize_head_pose, visualize_landmarks, visualize_mask
 
 def process_visibility(visibility_output):
     probs = torch.sigmoid(visibility_output[0])
@@ -81,15 +81,11 @@ def process_landmarks(landmark_output,images ):
     landmarks_list = []
     for landmark in denorm_landmarks[0]:
         x, y = landmark[0], landmark[1]
-        landmarks_list.append((int(x.item()), int(y.item())))
+        # landmarks_list.append((int(x.item()), int(y.item())))
+        landmarks_list.append((int(round(x.item())), int(round(y.item()))))
 
-    # landmarks_dict = {}
-    # for index, landmark in enumerate(denorm_landmarks[0]):
-    #     x, y = landmark[0], landmark[1]
-    #     landmarks_dict[f"landmark_{index}"] = (x.item(), y.item())
-
-    im = visualize_landmarks(image, denorm_landmarks, (255, 255, 0))
+    # im = visualize_landmarks(image, denorm_landmarks, (255, 255, 0))
     # save_path_viz = os.path.join(args.results_path, "landmarks.png")
-    cv2.imwrite("./landmarks.png", im[:, :, ::-1])
+    # cv2.imwrite("./landmarks.png", im[:, :, ::-1])
 
     return landmarks_list

setup.py

@@ -6,7 +6,7 @@
 
 setup(
     name='facexformer_pipeline',  # Package name
-    version='0.2.1',  # Version of your package
+    version='0.2.2',  # Version of your package
     author='Enes Kuzucu',  # Your name
 
     description='A module to run facexformer model as pipeline',  # Short description