Spaces:
Running
on
Zero
Running
on
Zero
moved resizing to 3D model code out of depth generatino to clean architecture
Browse files
app.py
CHANGED
@@ -60,7 +60,7 @@ def generate_3d_model(depth, image_path, focallength_px):
|
|
60 |
|
61 |
Args:
|
62 |
depth (np.ndarray): 2D array representing depth in meters.
|
63 |
-
image_path (str): Path to the
|
64 |
focallength_px (float): Focal length in pixels.
|
65 |
|
66 |
Returns:
|
@@ -68,8 +68,16 @@ def generate_3d_model(depth, image_path, focallength_px):
|
|
68 |
"""
|
69 |
# Load the RGB image and convert to a NumPy array
|
70 |
image = np.array(Image.open(image_path))
|
|
|
|
|
|
|
|
|
|
|
71 |
height, width = depth.shape
|
72 |
|
|
|
|
|
|
|
73 |
# Compute camera intrinsic parameters
|
74 |
fx = fy = focallength_px # Assuming square pixels and fx = fy
|
75 |
cx, cy = width / 2, height / 2 # Principal point at the image center
|
@@ -126,17 +134,13 @@ def predict_depth(input_image):
|
|
126 |
# Preprocess the image for depth prediction
|
127 |
result = depth_pro.load_rgb(temp_file)
|
128 |
|
129 |
-
# Add error checking for the result tuple
|
130 |
if len(result) < 2:
|
131 |
raise ValueError(f"Unexpected result from load_rgb: {result}")
|
132 |
|
133 |
-
image
|
134 |
-
f_px = result[-1] # Extract focal length
|
135 |
-
|
136 |
print(f"Extracted focal length: {f_px}")
|
137 |
|
138 |
-
image = transform(image)
|
139 |
-
image = image.to(device) # Move the image tensor to the selected device
|
140 |
|
141 |
# Run the depth prediction model
|
142 |
prediction = model.infer(image, f_px=f_px)
|
@@ -151,33 +155,13 @@ def predict_depth(input_image):
|
|
151 |
if depth.ndim != 2:
|
152 |
depth = depth.squeeze()
|
153 |
|
154 |
-
|
155 |
-
print(f"Original depth shape: {depth.shape}")
|
156 |
-
print(f"Original image shape: {image.shape}")
|
157 |
-
|
158 |
-
# Resize depth to match image dimensions
|
159 |
-
image_height, image_width = image.shape[2], image.shape[3]
|
160 |
-
depth = cv2.resize(depth, (image_width, image_height), interpolation=cv2.INTER_LINEAR)
|
161 |
-
|
162 |
-
print(f"Resized depth shape: {depth.shape}")
|
163 |
-
print(f"Final image shape: {image.shape}")
|
164 |
-
|
165 |
-
# No downsampling
|
166 |
-
downscale_factor = 1
|
167 |
-
|
168 |
-
# Convert image tensor to CPU and NumPy
|
169 |
-
image_np = image.cpu().detach().numpy()[0].transpose(1, 2, 0)
|
170 |
-
|
171 |
-
# No normalization of depth map as it is already in meters
|
172 |
-
depth_min = np.min(depth)
|
173 |
-
depth_max = np.max(depth)
|
174 |
-
depth_normalized = depth # Depth remains in meters
|
175 |
|
176 |
# Create a color map for visualization using matplotlib
|
177 |
plt.figure(figsize=(10, 10))
|
178 |
-
plt.imshow(
|
179 |
plt.colorbar(label='Depth [m]')
|
180 |
-
plt.title(f'Predicted Depth Map - Min: {
|
181 |
plt.axis('off') # Hide axis for a cleaner image
|
182 |
|
183 |
# Save the depth map visualization to a file
|
@@ -208,8 +192,9 @@ def get_last_commit_timestamp():
|
|
208 |
try:
|
209 |
timestamp = subprocess.check_output(['git', 'log', '-1', '--format=%cd', '--date=iso']).decode('utf-8').strip()
|
210 |
return datetime.fromisoformat(timestamp).strftime("%Y-%m-%d %H:%M:%S")
|
211 |
-
except Exception:
|
212 |
-
|
|
|
213 |
|
214 |
# Create the Gradio interface with appropriate input and output components.
|
215 |
last_updated = get_last_commit_timestamp()
|
|
|
60 |
|
61 |
Args:
|
62 |
depth (np.ndarray): 2D array representing depth in meters.
|
63 |
+
image_path (str): Path to the RGB image.
|
64 |
focallength_px (float): Focal length in pixels.
|
65 |
|
66 |
Returns:
|
|
|
68 |
"""
|
69 |
# Load the RGB image and convert to a NumPy array
|
70 |
image = np.array(Image.open(image_path))
|
71 |
+
|
72 |
+
# Resize depth to match image dimensions if necessary
|
73 |
+
if depth.shape != image.shape[:2]:
|
74 |
+
depth = cv2.resize(depth, (image.shape[1], image.shape[0]), interpolation=cv2.INTER_LINEAR)
|
75 |
+
|
76 |
height, width = depth.shape
|
77 |
|
78 |
+
print(f"3D model generation - Depth shape: {depth.shape}")
|
79 |
+
print(f"3D model generation - Image shape: {image.shape}")
|
80 |
+
|
81 |
# Compute camera intrinsic parameters
|
82 |
fx = fy = focallength_px # Assuming square pixels and fx = fy
|
83 |
cx, cy = width / 2, height / 2 # Principal point at the image center
|
|
|
134 |
# Preprocess the image for depth prediction
|
135 |
result = depth_pro.load_rgb(temp_file)
|
136 |
|
|
|
137 |
if len(result) < 2:
|
138 |
raise ValueError(f"Unexpected result from load_rgb: {result}")
|
139 |
|
140 |
+
image, _, _, _, f_px = result
|
|
|
|
|
141 |
print(f"Extracted focal length: {f_px}")
|
142 |
|
143 |
+
image = transform(image).to(device)
|
|
|
144 |
|
145 |
# Run the depth prediction model
|
146 |
prediction = model.infer(image, f_px=f_px)
|
|
|
155 |
if depth.ndim != 2:
|
156 |
depth = depth.squeeze()
|
157 |
|
158 |
+
print(f"Depth map shape: {depth.shape}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
159 |
|
160 |
# Create a color map for visualization using matplotlib
|
161 |
plt.figure(figsize=(10, 10))
|
162 |
+
plt.imshow(depth, cmap='gist_rainbow')
|
163 |
plt.colorbar(label='Depth [m]')
|
164 |
+
plt.title(f'Predicted Depth Map - Min: {np.min(depth):.1f}m, Max: {np.max(depth):.1f}m')
|
165 |
plt.axis('off') # Hide axis for a cleaner image
|
166 |
|
167 |
# Save the depth map visualization to a file
|
|
|
192 |
try:
|
193 |
timestamp = subprocess.check_output(['git', 'log', '-1', '--format=%cd', '--date=iso']).decode('utf-8').strip()
|
194 |
return datetime.fromisoformat(timestamp).strftime("%Y-%m-%d %H:%M:%S")
|
195 |
+
except Exception as e:
|
196 |
+
print(f"{str(e)}")
|
197 |
+
return str(e)
|
198 |
|
199 |
# Create the Gradio interface with appropriate input and output components.
|
200 |
last_updated = get_last_commit_timestamp()
|