Update app.py

app.py

@@ -1,79 +1,93 @@
 import streamlit as st
+import cv2
 import open3d as o3d
 import numpy as np
 import tempfile
 import os
-from PIL import Image
 
 # Title of the App
-st.title("3D Reconstruction Tool from Images 📷 → 🛠️ → 🧊")
+st.title("3D Reconstruction Tool from Video 📹 → 🛠️ → 🧊")
 
 # Sidebar: Information
 st.sidebar.write("""
 ## About the App
-Upload multiple images of an object taken from different angles.  
-The app will reconstruct a 3D point cloud, generate a surface mesh, and allow you to visualize and download the 3D model.
+Upload a video file, extract frames, reconstruct a 3D point cloud using Structure from Motion (SfM), and visualize or download the 3D mesh.
 """)
 
-# Step 1: Upload Images
-uploaded_images = st.file_uploader(
-    "Upload Multiple Images (PNG, JPG, JPEG)", 
-    type=["png", "jpg", "jpeg"], 
-    accept_multiple_files=True
-)
-
-# Check if images are uploaded
-if uploaded_images:
-    st.write(f"✅ {len(uploaded_images)} images uploaded successfully!")
-
-    # Step 2: Process Uploaded Images
-    st.write("🔄 Generating 3D Point Cloud from the images...")
-
-    # Save uploaded images temporarily
-    images_dir = tempfile.mkdtemp()
-    image_paths = []
-    for idx, uploaded_file in enumerate(uploaded_images):
-        image_path = os.path.join(images_dir, f"image_{idx:04d}.png")
-        image = Image.open(uploaded_file)
-        image.save(image_path)
-        image_paths.append(image_path)
-
-    # Step 3: Generate Point Cloud (Simplified Approach)
-    # Combine grayscale data into a pseudo-3D point cloud
+# Step 1: Upload Video File
+uploaded_file = st.file_uploader("Upload a Video File (MP4, AVI)", type=["mp4", "avi"])
+
+# Function to extract frames from video
+def extract_frames(video_path, frame_rate=10):
+    cap = cv2.VideoCapture(video_path)
+    frames = []
+    count = 0
+    while cap.isOpened():
+        ret, frame = cap.read()
+        if not ret:
+            break
+        if count % frame_rate == 0:
+            frames.append(frame)
+        count += 1
+    cap.release()
+    return frames
+
+# Function to save frames as images
+def save_frames_as_images(frames, output_dir):
+    os.makedirs(output_dir, exist_ok=True)
+    for i, frame in enumerate(frames):
+        filename = os.path.join(output_dir, f"frame_{i:04d}.png")
+        cv2.imwrite(filename, frame)
+
+# Step 2: Process Uploaded Video
+if uploaded_file:
+    st.video(uploaded_file)
+    st.write("Extracting frames...")
+
+    # Save the uploaded video temporarily
+    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp4") as tmp_video:
+        tmp_video.write(uploaded_file.read())
+        video_path = tmp_video.name
+
+    # Extract frames
+    frames = extract_frames(video_path, frame_rate=10)
+    st.write(f"✅ Extracted {len(frames)} frames from the video.")
+
+    # Save extracted frames
+    frames_dir = tempfile.mkdtemp()
+    save_frames_as_images(frames, frames_dir)
+    st.write(f"Frames saved temporarily at `{frames_dir}`.")
+
+    # Step 3: Structure from Motion (3D Reconstruction)
+    st.write("🔄 Reconstructing 3D point cloud using Structure from Motion...")
+    
+    # Create Open3D Point Cloud
     pcd = o3d.geometry.PointCloud()
-
-    for image_path in image_paths:
-        # Load image and convert to grayscale
-        img = np.array(Image.open(image_path).convert("L"))
-        height, width = img.shape
-
-        # Generate x, y, z coordinates (z = intensity as pseudo-depth)
+    for image_file in sorted(os.listdir(frames_dir)):
+        img_path = os.path.join(frames_dir, image_file)
+        frame = cv2.imread(img_path)
+        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
+        
+        # Dummy point cloud generation for simplicity
+        height, width = gray.shape
         x, y = np.meshgrid(np.arange(width), np.arange(height))
-        z = img / 255.0  # Normalize grayscale intensity to pseudo-depth
+        z = gray / 255.0  # Use gray intensity as a pseudo depth
         points = np.stack((x.flatten(), y.flatten(), z.flatten()), axis=1)
-
-        # Add points to the point cloud
         pcd.points.extend(o3d.utility.Vector3dVector(points))
 
-    # Step 4: Surface Reconstruction using Poisson Reconstruction
-    st.write("🛠️ Generating mesh using Poisson Surface Reconstruction...")
-
-    # Estimate normals for the point cloud
+    # Step 4: Surface Reconstruction
+    st.write("🛠️ Generating mesh using Poisson Reconstruction...")
     pcd.estimate_normals()
     mesh, _ = o3d.geometry.TriangleMesh.create_from_point_cloud_poisson(pcd, depth=8)
 
     # Step 5: Visualization
     st.write("✅ Reconstruction Complete! Visualizing the 3D model:")
+    o3d.io.write_triangle_mesh("reconstructed_mesh.stl", mesh)
 
-    # Save the mesh temporarily for visualization
-    output_mesh_path = "reconstructed_mesh.stl"
-    o3d.io.write_triangle_mesh(output_mesh_path, mesh)
-
-    # Use Plotly to visualize the 3D mesh
+    # Use Plotly for visualization
     import plotly.graph_objects as go
     vertices = np.asarray(mesh.vertices)
     triangles = np.asarray(mesh.triangles)
-
     fig = go.Figure(data=[go.Mesh3d(
         x=vertices[:, 0],
         y=vertices[:, 1],
@@ -88,5 +102,5 @@ if uploaded_images:
 
     # Step 6: Download the Optimized Mesh
     st.write("📥 Download the reconstructed 3D model:")
-    with open(output_mesh_path, "rb") as f:
-        st.download_button("Download 3D Mesh (STL)", f, file_name="3D_Model.stl")
+    with open("reconstructed_mesh.stl", "rb") as f:
+        st.download_button("Download 3D Mesh (STL)", f, file_name="reconstructed_3D_Model.stl")