Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -35,12 +35,15 @@ pipe.unet.eval()
|
|
35 |
|
36 |
# UI texts
|
37 |
title = "# End-to-End Fine-Tuned GeoWizard Video"
|
38 |
-
description = """
|
|
|
|
|
39 |
|
40 |
@spaces.GPU
|
41 |
def predict(image: Image.Image, processing_res_choice: int):
|
42 |
"""
|
43 |
Single-frame prediction wrapped for GPU execution.
|
|
|
44 |
"""
|
45 |
with torch.no_grad():
|
46 |
return pipe(
|
@@ -67,7 +70,7 @@ def on_submit_video(video_path: str, processing_res_choice: int):
|
|
67 |
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
68 |
frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
69 |
|
70 |
-
#
|
71 |
tmp_depth = tempfile.NamedTemporaryFile(suffix='.mp4', delete=False)
|
72 |
tmp_normal = tempfile.NamedTemporaryFile(suffix='.mp4', delete=False)
|
73 |
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
|
@@ -80,13 +83,14 @@ def on_submit_video(video_path: str, processing_res_choice: int):
|
|
80 |
if not ret:
|
81 |
break
|
82 |
|
83 |
-
# Convert BGR to RGB PIL
|
84 |
rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
85 |
pil_image = Image.fromarray(rgb)
|
86 |
|
87 |
# Run prediction
|
88 |
-
|
89 |
-
|
|
|
90 |
|
91 |
# Write depth frame
|
92 |
depth_frame = np.array(depth_colored)
|
@@ -103,6 +107,7 @@ time_error
|
|
103 |
out_depth.release()
|
104 |
out_normal.release()
|
105 |
|
|
|
106 |
return tmp_depth.name, tmp_normal.name
|
107 |
|
108 |
# Build Gradio interface
|
|
|
35 |
|
36 |
# UI texts
|
37 |
title = "# End-to-End Fine-Tuned GeoWizard Video"
|
38 |
+
description = """
|
39 |
+
Please refer to our [paper](https://arxiv.org/abs/2409.11355) and [GitHub](https://vision.rwth-aachen.de/diffusion-e2e-ft) for more details.
|
40 |
+
"""
|
41 |
|
42 |
@spaces.GPU
|
43 |
def predict(image: Image.Image, processing_res_choice: int):
|
44 |
"""
|
45 |
Single-frame prediction wrapped for GPU execution.
|
46 |
+
Returns a DepthNormalPipelineOutput with attributes depth_colored and normal_colored.
|
47 |
"""
|
48 |
with torch.no_grad():
|
49 |
return pipe(
|
|
|
70 |
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
71 |
frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
72 |
|
73 |
+
# Temporary output files
|
74 |
tmp_depth = tempfile.NamedTemporaryFile(suffix='.mp4', delete=False)
|
75 |
tmp_normal = tempfile.NamedTemporaryFile(suffix='.mp4', delete=False)
|
76 |
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
|
|
|
83 |
if not ret:
|
84 |
break
|
85 |
|
86 |
+
# Convert BGR to RGB and to PIL
|
87 |
rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
88 |
pil_image = Image.fromarray(rgb)
|
89 |
|
90 |
# Run prediction
|
91 |
+
result = predict(pil_image, processing_res_choice)
|
92 |
+
depth_colored = result.depth_colored
|
93 |
+
normal_colored = result.normal_colored
|
94 |
|
95 |
# Write depth frame
|
96 |
depth_frame = np.array(depth_colored)
|
|
|
107 |
out_depth.release()
|
108 |
out_normal.release()
|
109 |
|
110 |
+
# Return paths for download
|
111 |
return tmp_depth.name, tmp_normal.name
|
112 |
|
113 |
# Build Gradio interface
|