Update app.py

app.py

@@ -22,96 +22,37 @@ except Exception as e:
     print(f"Error installing detectron2: {e}")
     sys.exit(1)
 
-# Create a custom distributed.py file that doesn't need mpi4py
-os.makedirs('utils', exist_ok=True)
-with open('utils/distributed.py', 'w') as f:
-    f.write("""# Custom distributed.py without mpi4py dependency
-import os
-import torch
-import torch.distributed as dist
-
-class MPI:
-    class COMM_WORLD:
-        @staticmethod
-        def Get_rank():
-            return 0
-        @staticmethod
-        def Get_size():
-            return 1
-
-def init_distributed(opt=None):
-    if opt is not None:
-        opt.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-        opt.rank = 0
-        opt.world_size = 1
-        opt.gpu = 0
-        return opt
-    
-    return None
-
-def get_rank():
-    return 0
-
-def get_world_size():
-    return 1
-
-def is_main_process():
-    return True
-
-def synchronize():
-    pass
-
-def all_gather(data):
-    return [data]
-
-def reduce_dict(input_dict, average=True):
-    return input_dict
-""")
-
-# Create a simple visualizer if it doesn't exist
-if not os.path.exists('utils/visualizer.py'):
-    with open('utils/visualizer.py', 'w') as f:
-        f.write("""# Simple visualizer class
-import numpy as np
-import cv2
-
-class Visualizer:
-    def __init__(self, img_rgb, metadata=None, scale=1.0):
-        self.img = img_rgb
-        self.metadata = metadata
-        self.scale = scale
-
-    def draw_binary_mask(self, mask, color=None, text=None):
-        if color is None:
-            color = [0, 255, 0]  # Default to green
-        
-        mask_img = np.zeros_like(self.img, dtype=np.float32)
-        color_mask = np.array(color) * 255
-        
-        for c in range(3):
-            mask_img[:, :, c] = color_mask[c]
-        
-        mask_img = mask_img * mask[:, :, None] * 0.5
-        self.img = self.img * (1 - mask[:, :, None] * 0.5) + mask_img
-        
-        if text:
-            # Simplified text placement
-            x, y = np.where(mask)[0][0], np.where(mask)[1][0] if np.any(mask) else (10, 10)
-            cv2.putText(self.img, text, (y, x), cv2.FONT_HERSHEY_SIMPLEX, 0.5, tuple(map(int, color_mask)), 1)
-        
-        return self
-    
-    def draw_panoptic_seg(self, panoptic_seg, segments_info):
-        # Simplified panoptic visualization - just a placeholder
-        return self
-
-    def get_image(self):
-        return self.img.astype(np.uint8)
-""")
-
-# Set Python path to include the repository root
-os.environ["PYTHONPATH"] = os.getcwd()
-print(f"Set PYTHONPATH to: {os.getcwd()}")
+# Fix the distributed.py file if it's causing issues
+if os.path.exists('utils/distributed.py'):
+    with open('utils/distributed.py', 'r') as f:
+        content = f.read()
+    if 'from mpi4py import MPI' in content:
+        print("Patching utils/distributed.py to work without mpi4py")
+        patched_content = content.replace(
+            "from mpi4py import MPI",
+            """try:
+    from mpi4py import MPI
+except ImportError:
+    # Dummy MPI implementation
+    class MPI:
+        class COMM_WORLD:
+            @staticmethod
+            def Get_rank():
+                return 0
+            @staticmethod
+            def Get_size():
+                return 1"""
+        )
+        with open('utils/distributed.py', 'w') as f:
+            f.write(patched_content)
+        print("Patched utils/distributed.py")
+
+# Ensure the Python path includes the current directory
+current_dir = os.getcwd()
+if current_dir not in sys.path:
+    sys.path.insert(0, current_dir)
+os.environ["PYTHONPATH"] = current_dir
+print(f"Set PYTHONPATH to: {current_dir}")
 
 # Continue with regular imports
 import warnings
@@ -132,8 +73,22 @@ from utils.distributed import init_distributed
 from utils.arguments import load_opt_from_config_files
 from utils.constants import COCO_PANOPTIC_CLASSES
 
-# Import the interactive functions from the existing implementation
-from demo.seem.tasks.interactive import interactive_infer_image, interactive_infer_video
+# Import the interactive functions using a try-except block to catch import errors
+try:
+    from demo.seem.tasks.interactive import interactive_infer_image, interactive_infer_video
+    print("Successfully imported interactive functions")
+except ImportError as e:
+    print(f"Error importing interactive functions: {e}")
+    print("Python path:", sys.path)
+    print("Current directory:", os.getcwd())
+    print("Contents of current directory:", os.listdir('.'))
+    if os.path.exists('demo'):
+        print("Contents of demo directory:", os.listdir('demo'))
+        if os.path.exists('demo/seem'):
+            print("Contents of demo/seem directory:", os.listdir('demo/seem'))
+            if os.path.exists('demo/seem/tasks'):
+                print("Contents of demo/seem/tasks directory:", os.listdir('demo/seem/tasks'))
+    sys.exit(1)
 
 def parse_option():
     parser = argparse.ArgumentParser('SEEM Demo', add_help=False)
@@ -215,10 +170,20 @@ def inference(image, task, *args, **kwargs):
     video_pth = kwargs.get("video", None)
     
     # Call the appropriate interactive function
-    if 'Video' in task:
-        return interactive_infer_video(model, audio, image_input, task, refimg, reftxt, audio_pth, video_pth)
-    else:
-        return interactive_infer_image(model, audio, image_input, task, refimg, reftxt, audio_pth, video_pth)
+    try:
+        if 'Video' in task:
+            return interactive_infer_video(model, audio, image_input, task, refimg, reftxt, audio_pth, video_pth)
+        else:
+            return interactive_infer_image(model, audio, image_input, task, refimg, reftxt, audio_pth, video_pth)
+    except Exception as e:
+        print(f"Error during inference: {e}")
+        import traceback
+        traceback.print_exc()
+        warning_img = Image.new('RGB', (600, 400), color=(240, 240, 240))
+        d = ImageDraw.Draw(warning_img)
+        d.text((50, 150), f"Error: {str(e)}", fill=(255, 0, 0))
+        d.text((50, 200), "Please check logs for details.", fill=(255, 0, 0))
+        return warning_img, None
 
 class ImageMask(gr.components.Image):
     """