Spaces:
Running
on
Zero
Running
on
Zero
more debugging to address possible GPU OOM or timeout for 3D generation
Browse files
app.py
CHANGED
@@ -70,95 +70,106 @@ def resize_image(image_path, max_size=1024):
|
|
70 |
img.save(temp_file, format="PNG")
|
71 |
return temp_file.name
|
72 |
|
73 |
-
@spaces.GPU(duration=
|
74 |
def generate_3d_model(depth, image_path, focallength_px, simplification_factor=0.8, smoothing_iterations=1, thin_threshold=0.01):
|
75 |
"""
|
76 |
Generate a textured 3D mesh from the depth map and the original image.
|
77 |
"""
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
depth
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
print("
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
162 |
|
163 |
def remove_thin_features(mesh, thickness_threshold=0.01):
|
164 |
"""
|
|
|
70 |
img.save(temp_file, format="PNG")
|
71 |
return temp_file.name
|
72 |
|
73 |
+
@spaces.GPU(duration=30) # Increased duration to 30 seconds
|
74 |
def generate_3d_model(depth, image_path, focallength_px, simplification_factor=0.8, smoothing_iterations=1, thin_threshold=0.01):
|
75 |
"""
|
76 |
Generate a textured 3D mesh from the depth map and the original image.
|
77 |
"""
|
78 |
+
try:
|
79 |
+
print("Starting 3D model generation")
|
80 |
+
# Load the RGB image and convert to a NumPy array
|
81 |
+
image = np.array(Image.open(image_path))
|
82 |
+
|
83 |
+
# Ensure depth is a NumPy array
|
84 |
+
if isinstance(depth, torch.Tensor):
|
85 |
+
depth = depth.cpu().numpy()
|
86 |
+
|
87 |
+
# Resize depth to match image dimensions if necessary
|
88 |
+
if depth.shape != image.shape[:2]:
|
89 |
+
depth = cv2.resize(depth, (image.shape[1], image.shape[0]), interpolation=cv2.INTER_LINEAR)
|
90 |
+
|
91 |
+
height, width = depth.shape
|
92 |
+
|
93 |
+
print(f"3D model generation - Depth shape: {depth.shape}")
|
94 |
+
print(f"3D model generation - Image shape: {image.shape}")
|
95 |
+
|
96 |
+
# Compute camera intrinsic parameters
|
97 |
+
fx = fy = float(focallength_px) # Ensure focallength_px is a float
|
98 |
+
cx, cy = width / 2, height / 2 # Principal point at the image center
|
99 |
+
|
100 |
+
# Create a grid of (u, v) pixel coordinates
|
101 |
+
u = np.arange(0, width)
|
102 |
+
v = np.arange(0, height)
|
103 |
+
uu, vv = np.meshgrid(u, v)
|
104 |
+
|
105 |
+
# Convert pixel coordinates to real-world 3D coordinates using the pinhole camera model
|
106 |
+
Z = depth.flatten()
|
107 |
+
X = ((uu.flatten() - cx) * Z) / fx
|
108 |
+
Y = ((vv.flatten() - cy) * Z) / fy
|
109 |
+
|
110 |
+
# Stack the coordinates to form vertices (X, Y, Z)
|
111 |
+
vertices = np.vstack((X, Y, Z)).T
|
112 |
+
|
113 |
+
# Normalize RGB colors to [0, 1] for vertex coloring
|
114 |
+
colors = image.reshape(-1, 3) / 255.0
|
115 |
+
|
116 |
+
print("Generating faces")
|
117 |
+
# Generate faces by connecting adjacent vertices to form triangles
|
118 |
+
faces = []
|
119 |
+
for i in range(height - 1):
|
120 |
+
for j in range(width - 1):
|
121 |
+
idx = i * width + j
|
122 |
+
# Triangle 1
|
123 |
+
faces.append([idx, idx + width, idx + 1])
|
124 |
+
# Triangle 2
|
125 |
+
faces.append([idx + 1, idx + width, idx + width + 1])
|
126 |
+
faces = np.array(faces)
|
127 |
+
|
128 |
+
print("Creating mesh")
|
129 |
+
# Create the mesh using Trimesh with vertex colors
|
130 |
+
mesh = trimesh.Trimesh(vertices=vertices, faces=faces, vertex_colors=colors)
|
131 |
+
|
132 |
+
# Mesh cleaning and improvement steps
|
133 |
+
print("Original mesh - vertices: {}, faces: {}".format(len(mesh.vertices), len(mesh.faces)))
|
134 |
+
|
135 |
+
print("Simplifying mesh")
|
136 |
+
# 1. Mesh simplification
|
137 |
+
target_faces = int(len(mesh.faces) * simplification_factor)
|
138 |
+
mesh = mesh.simplify_quadric_decimation(face_count=target_faces)
|
139 |
+
print("After simplification - vertices: {}, faces: {}".format(len(mesh.vertices), len(mesh.faces)))
|
140 |
+
|
141 |
+
print("Removing small components")
|
142 |
+
# 2. Remove small disconnected components
|
143 |
+
components = mesh.split(only_watertight=False)
|
144 |
+
if len(components) > 1:
|
145 |
+
areas = np.array([c.area for c in components])
|
146 |
+
mesh = components[np.argmax(areas)]
|
147 |
+
print("After removing small components - vertices: {}, faces: {}".format(len(mesh.vertices), len(mesh.faces)))
|
148 |
+
|
149 |
+
print("Smoothing mesh")
|
150 |
+
# 3. Smooth the mesh
|
151 |
+
for _ in range(smoothing_iterations):
|
152 |
+
mesh = mesh.smoothed()
|
153 |
+
print("After smoothing - vertices: {}, faces: {}".format(len(mesh.vertices), len(mesh.faces)))
|
154 |
+
|
155 |
+
print("Removing thin features")
|
156 |
+
# 4. Remove thin features
|
157 |
+
mesh = remove_thin_features(mesh, thickness_threshold=thin_threshold)
|
158 |
+
print("After removing thin features - vertices: {}, faces: {}".format(len(mesh.vertices), len(mesh.faces)))
|
159 |
+
|
160 |
+
# Export the mesh to OBJ files with unique filenames
|
161 |
+
timestamp = int(time.time())
|
162 |
+
view_model_path = f'view_model_{timestamp}.obj'
|
163 |
+
download_model_path = f'download_model_{timestamp}.obj'
|
164 |
+
print("Exporting to view")
|
165 |
+
mesh.export(view_model_path)
|
166 |
+
print("Exporting to download")
|
167 |
+
mesh.export(download_model_path)
|
168 |
+
print("Export completed")
|
169 |
+
return view_model_path, download_model_path
|
170 |
+
except Exception as e:
|
171 |
+
print(f"Error in generate_3d_model: {str(e)}")
|
172 |
+
raise
|
173 |
|
174 |
def remove_thin_features(mesh, thickness_threshold=0.01):
|
175 |
"""
|