test

anydoor/configs/inference.yaml

@@ -1,3 +1,3 @@
-pretrained_model: /work/wefa-door-master/adbase-step=48745.ckpt
+pretrained_model: /work/wefa-door-master/adbase-step=41250.ckpt
 config_file: configs/anydoor.yaml
 save_memory: False

anydoor/run_inference_train_x.py

@@ -156,7 +156,7 @@ def inference_single_image(ref_image, ref_mask, tar_image, tar_mask, guidance_sc
     hint_image = hint[:,:,:-1]
     hint_mask = item['hint'][:,:,-1] * 255
     hint_mask = np.stack([hint_mask,hint_mask,hint_mask],-1)
-    ref = cv2.resize(ref.astype(np.uint8), (512,512))
+    ref = cv2.resize(ref.astype(np.uint8), (512,682))
 
     seed = random.randint(0, 65535)
     if save_memory:
@@ -186,15 +186,14 @@ def inference_single_image(ref_image, ref_mask, tar_image, tar_mask, guidance_sc
     if save_memory:
         model.low_vram_shift(is_diffusing=True)
 
-    # ====
     num_samples = 1 #gr.Slider(label="Images", minimum=1, maximum=12, value=1, step=1)
-    image_resolution = 512  #gr.Slider(label="Image Resolution", minimum=256, maximum=768, value=512, step=64)
+    image_resolution = 768  #gr.Slider(label="Image Resolution", minimum=256, maximum=768, value=512, step=64)
     strength = 1  #gr.Slider(label="Control Strength", minimum=0.0, maximum=2.0, value=1.0, step=0.01)
     guess_mode = False #gr.Checkbox(label='Guess Mode', value=False)
     #detect_resolution = 512  #gr.Slider(label="Segmentation Resolution", minimum=128, maximum=1024, value=512, step=1)
     ddim_steps = 50 #gr.Slider(label="Steps", minimum=1, maximum=100, value=20, step=1)
     scale = guidance_scale  #gr.Slider(label="Guidance Scale", minimum=0.1, maximum=30.0, value=9.0, step=0.1)
-    seed = -1  #gr.Slider(label="Seed", minimum=-1, maximum=2147483647, step=1, randomize=True)
+    seed = 135  #gr.Slider(label="Seed", minimum=-1, maximum=2147483647, step=1, randomize=True)
     eta = 0.0 #gr.Number(label="eta (DDIM)", value=0.0)
 
     model.control_scales = [strength * (0.825 ** float(12 - i)) for i in range(13)] if guess_mode else ([strength] * 13)  # Magic number. IDK why. Perhaps because 0.825**12<0.01 but 0.826**12>0.01
@@ -219,6 +218,20 @@ def inference_single_image(ref_image, ref_mask, tar_image, tar_mask, guidance_sc
     return gen_image
 
 
+def resize_and_crop(image, target_width, target_height, top_offset, bottom_offset):
+    # Crop the center horizontally
+    center_x = image.shape[1] // 2
+    half_width = target_width // 2
+    cropped_image = image[:, center_x - half_width:center_x + half_width]
+    
+    # Crop the top or bottom part vertically
+    if top_offset:
+        cropped_image = cropped_image[:target_height]
+    elif bottom_offset:
+        cropped_image = cropped_image[-target_height:]
+    
+    return cropped_image
+
 if __name__ == '__main__': 
 
     import os
@@ -236,35 +249,74 @@ if __name__ == '__main__':
     # Fixed reference image and its mask
     fixed_ref_image_name = 'we_picked_8.jpg'
     fixed_ref_image_path = os.path.join(image_dir, fixed_ref_image_name)
-    fixed_ref_mask_path = os.path.join(image_parse_v3_dir,  'top_long_collar.png')
+    fixed_ref_mask_path = os.path.join(image_parse_v3_dir,  'T_we_picked_8.png')
 
     # Ensure the save directory exists
     if not os.path.exists(save_dir):
         os.makedirs(save_dir)
 
     # Create list of cloth image names
-    cloth_image_names = os.listdir(cloth_dir)
+    cloth_image_names = os.listdir(cloth_dir)[:5]
+
+    # First loop for the top half
+    for cloth_image_name in cloth_image_names:
+        cloth_image_path = os.path.join(cloth_dir, cloth_image_name)
+        cloth_mask_path = os.path.join(cloth_mask_dir, cloth_image_name)
+        
+        cloth_image = cv2.imread(cloth_image_path)
+        cloth_image = cv2.cvtColor(cloth_image, cv2.COLOR_BGR2RGB)
+        cloth_mask = (cv2.imread(cloth_mask_path) > 128).astype(np.uint8)[:, :, 0]
+
+        ref_image = cv2.imread(fixed_ref_image_path)
+        ref_image = cv2.cvtColor(ref_image, cv2.COLOR_BGR2RGB)
+        ref_mask = Image.open(fixed_ref_mask_path).convert('P')
+        ref_mask = np.array(ref_mask) == 1  
+
+        gen_image = inference_single_image(cloth_image, cloth_mask, ref_image, ref_mask)
+        gen_path = os.path.join(save_dir, 'gen_top_' + cloth_image_name)
+        single_path = os.path.join(save_dir, 'single_top_' + cloth_image_name)
+        clothe_path = os.path.join(save_dir, 'single_cloth_top_' + cloth_image_name)
 
+        vis_image = cv2.hconcat([cloth_image, ref_image, gen_image])
+        cv2.imwrite(gen_path, cv2.cvtColor(vis_image, cv2.COLOR_RGB2BGR))
+        cv2.imwrite(clothe_path, cv2.cvtColor(cloth_image, cv2.COLOR_RGB2BGR))
+        cv2.imwrite(single_path, cv2.cvtColor(gen_image, cv2.COLOR_RGB2BGR))
+        top_image_path = os.path.join(save_dir, 'top_z_' + cloth_image_name)
+        top_half_cropped = resize_and_crop(gen_image, 512, 600, True, False)
+        cv2.imwrite(top_image_path, cv2.cvtColor(top_half_cropped, cv2.COLOR_RGB2BGR))
+
+
+
+    # Update the cloth directory and mask path for the bottom half processing
+    cloth_dir = '/work/pink_girl/cloth/trou'
+    fixed_ref_mask_path = os.path.join(image_parse_v3_dir, 'trou1.png')
+    cloth_image_names = os.listdir(cloth_dir)[:5]
+
+    # Second loop for the bottom half
     for cloth_image_name in cloth_image_names:
-        # Construct paths for cloth and its mask
         cloth_image_path = os.path.join(cloth_dir, cloth_image_name)
         cloth_mask_path = os.path.join(cloth_mask_dir, cloth_image_name)
         
-        # Load and process the cloth image and mask
         cloth_image = cv2.imread(cloth_image_path)
         cloth_image = cv2.cvtColor(cloth_image, cv2.COLOR_BGR2RGB)
         cloth_mask = (cv2.imread(cloth_mask_path) > 128).astype(np.uint8)[:, :, 0]
 
-        # Load and process the fixed reference image and mask
         ref_image = cv2.imread(fixed_ref_image_path)
         ref_image = cv2.cvtColor(ref_image, cv2.COLOR_BGR2RGB)
         ref_mask = Image.open(fixed_ref_mask_path).convert('P')
-        ref_mask = np.array(ref_mask) == 5  # Update this value if the mask classification is different
+        ref_mask = np.array(ref_mask) == 2  
 
-        # Generate the image using the provided function
         gen_image = inference_single_image(cloth_image, cloth_mask, ref_image, ref_mask)
-        gen_path = os.path.join(save_dir, '5_' + cloth_image_name)
+        gen_path = os.path.join(save_dir, 'gen_pant_' + cloth_image_name)
+        single_path = os.path.join(save_dir, 'single_pant_' + cloth_image_name)
+        clothe_path = os.path.join(save_dir, 'single_cloth_pant_' + cloth_image_name)
 
-        # Concatenate and save the visualization
         vis_image = cv2.hconcat([cloth_image, ref_image, gen_image])
         cv2.imwrite(gen_path, cv2.cvtColor(vis_image, cv2.COLOR_RGB2BGR))
+        cv2.imwrite(clothe_path, cv2.cvtColor(cloth_image, cv2.COLOR_RGB2BGR))
+        cv2.imwrite(single_path, cv2.cvtColor(gen_image, cv2.COLOR_RGB2BGR))
+
+        # Replace the resizing and padding code for the bottom half
+        bottom_image_path = os.path.join(save_dir, 'bottom_z_' + cloth_image_name)
+        bottom_half_cropped = resize_and_crop(gen_image, 512, 600, False, True)
+        cv2.imwrite(bottom_image_path, cv2.cvtColor(bottom_half_cropped, cv2.COLOR_RGB2BGR))