Update app.py
Browse files
app.py
CHANGED
@@ -13,20 +13,17 @@ import tempfile
|
|
13 |
import uuid
|
14 |
import time
|
15 |
import threading
|
|
|
16 |
|
17 |
torch.set_float32_matmul_precision("medium")
|
18 |
-
|
19 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
20 |
|
21 |
# Load both BiRefNet models
|
22 |
birefnet = AutoModelForImageSegmentation.from_pretrained(
|
23 |
-
"ZhengPeng7/BiRefNet", trust_remote_code=True
|
24 |
-
)
|
25 |
birefnet.to(device)
|
26 |
-
|
27 |
birefnet_lite = AutoModelForImageSegmentation.from_pretrained(
|
28 |
-
"ZhengPeng7/BiRefNet_lite", trust_remote_code=True
|
29 |
-
)
|
30 |
birefnet_lite.to(device)
|
31 |
|
32 |
transform_image = transforms.Compose(
|
@@ -37,7 +34,6 @@ transform_image = transforms.Compose(
|
|
37 |
]
|
38 |
)
|
39 |
|
40 |
-
|
41 |
# Function to delete files older than 10 minutes in the temp directory
|
42 |
def cleanup_temp_files():
|
43 |
while True:
|
@@ -55,11 +51,29 @@ def cleanup_temp_files():
|
|
55 |
print(f"Error deleting file {filepath}: {e}")
|
56 |
time.sleep(60) # Check every minute
|
57 |
|
58 |
-
|
59 |
# Start the cleanup thread
|
60 |
cleanup_thread = threading.Thread(target=cleanup_temp_files, daemon=True)
|
61 |
cleanup_thread.start()
|
62 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
63 |
|
64 |
@spaces.GPU
|
65 |
def fn(vid, bg_type="Color", bg_image=None, bg_video=None, color="#00FF00", fps=0, video_handling="slow_down", fast_mode=True):
|
@@ -77,7 +91,7 @@ def fn(vid, bg_type="Color", bg_image=None, bg_video=None, color="#00FF00", fps=
|
|
77 |
audio = video.audio
|
78 |
|
79 |
# Extract frames at the specified FPS
|
80 |
-
frames = video.iter_frames(fps=fps)
|
81 |
|
82 |
# Process each frame for background removal
|
83 |
processed_frames = []
|
@@ -96,29 +110,14 @@ def fn(vid, bg_type="Color", bg_image=None, bg_video=None, color="#00FF00", fps=
|
|
96 |
|
97 |
bg_frame_index = 0 # Initialize background frame index
|
98 |
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
background_frame = background_frames[bg_frame_index % len(background_frames)]
|
108 |
-
bg_frame_index += 1
|
109 |
-
background_image = Image.fromarray(background_frame)
|
110 |
-
processed_image = process(pil_image, background_image, fast_mode)
|
111 |
-
else: # video_handling == "loop"
|
112 |
-
background_frame = background_frames[bg_frame_index % len(background_frames)]
|
113 |
-
bg_frame_index += 1
|
114 |
-
background_image = Image.fromarray(background_frame)
|
115 |
-
processed_image = process(pil_image, background_image, fast_mode)
|
116 |
-
else:
|
117 |
-
processed_image = pil_image # Default to original image if no background is selected
|
118 |
-
|
119 |
-
processed_frames.append(np.array(processed_image))
|
120 |
-
elapsed_time = time.time() - start_time
|
121 |
-
yield processed_image, None, f"Processing frame {i+1}... Elapsed time: {elapsed_time:.2f} seconds"
|
122 |
|
123 |
# Create a new video from the processed frames
|
124 |
processed_video = mp.ImageSequenceClip(processed_frames, fps=fps)
|
@@ -135,8 +134,9 @@ def fn(vid, bg_type="Color", bg_image=None, bg_video=None, color="#00FF00", fps=
|
|
135 |
|
136 |
elapsed_time = time.time() - start_time
|
137 |
yield gr.update(visible=False), gr.update(visible=True), f"Processing complete! Elapsed time: {elapsed_time:.2f} seconds"
|
|
|
138 |
# Return the path to the temporary file
|
139 |
-
yield
|
140 |
|
141 |
except Exception as e:
|
142 |
print(f"Error: {e}")
|
@@ -144,7 +144,6 @@ def fn(vid, bg_type="Color", bg_image=None, bg_video=None, color="#00FF00", fps=
|
|
144 |
yield gr.update(visible=False), gr.update(visible=True), f"Error processing video: {e}. Elapsed time: {elapsed_time:.2f} seconds"
|
145 |
yield None, f"Error processing video: {e}", f"Error processing video: {e}. Elapsed time: {elapsed_time:.2f} seconds"
|
146 |
|
147 |
-
|
148 |
def process(image, bg, fast_mode=False):
|
149 |
image_size = image.size
|
150 |
input_images = transform_image(image).unsqueeze(0).to("cuda")
|
@@ -169,12 +168,10 @@ def process(image, bg, fast_mode=False):
|
|
169 |
|
170 |
# Composite the image onto the background using the mask
|
171 |
image = Image.composite(image, background, mask)
|
172 |
-
|
173 |
return image
|
174 |
|
175 |
-
|
176 |
with gr.Blocks(theme=gr.themes.Ocean()) as demo:
|
177 |
-
gr.Markdown("# Video Background Remover & Changer\n### You can replace image background with any color, image or video.\nNOTE: As this Space is running on ZERO GPU it has limit. It can handle approx
|
178 |
with gr.Row():
|
179 |
in_video = gr.Video(label="Input Video", interactive=True)
|
180 |
stream_image = gr.Image(label="Streaming Output", visible=False)
|
@@ -196,8 +193,7 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
|
|
196 |
with gr.Column(visible=False) as video_handling_options:
|
197 |
video_handling_radio = gr.Radio(["slow_down", "loop"], label="Video Handling", value="slow_down", interactive=True)
|
198 |
fast_mode_checkbox = gr.Checkbox(label="Fast Mode (Use BiRefNet_lite)", value=True, interactive=True)
|
199 |
-
|
200 |
-
time_textbox = gr.Textbox(label="Time Elapsed", interactive=False) # Add time textbox
|
201 |
|
202 |
def update_visibility(bg_type):
|
203 |
if bg_type == "Color":
|
|
|
13 |
import uuid
|
14 |
import time
|
15 |
import threading
|
16 |
+
from concurrent.futures import ThreadPoolExecutor
|
17 |
|
18 |
torch.set_float32_matmul_precision("medium")
|
|
|
19 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
20 |
|
21 |
# Load both BiRefNet models
|
22 |
birefnet = AutoModelForImageSegmentation.from_pretrained(
|
23 |
+
"ZhengPeng7/BiRefNet", trust_remote_code=True)
|
|
|
24 |
birefnet.to(device)
|
|
|
25 |
birefnet_lite = AutoModelForImageSegmentation.from_pretrained(
|
26 |
+
"ZhengPeng7/BiRefNet_lite", trust_remote_code=True)
|
|
|
27 |
birefnet_lite.to(device)
|
28 |
|
29 |
transform_image = transforms.Compose(
|
|
|
34 |
]
|
35 |
)
|
36 |
|
|
|
37 |
# Function to delete files older than 10 minutes in the temp directory
|
38 |
def cleanup_temp_files():
|
39 |
while True:
|
|
|
51 |
print(f"Error deleting file {filepath}: {e}")
|
52 |
time.sleep(60) # Check every minute
|
53 |
|
|
|
54 |
# Start the cleanup thread
|
55 |
cleanup_thread = threading.Thread(target=cleanup_temp_files, daemon=True)
|
56 |
cleanup_thread.start()
|
57 |
|
58 |
+
# Function to process a single frame
|
59 |
+
def process_frame(frame, bg_type, bg, fast_mode, bg_frame_index, background_frames, color):
|
60 |
+
try:
|
61 |
+
pil_image = Image.fromarray(frame)
|
62 |
+
if bg_type == "Color":
|
63 |
+
processed_image = process(pil_image, color, fast_mode)
|
64 |
+
elif bg_type == "Image":
|
65 |
+
processed_image = process(pil_image, bg, fast_mode)
|
66 |
+
elif bg_type == "Video":
|
67 |
+
background_frame = background_frames[bg_frame_index % len(background_frames)]
|
68 |
+
bg_frame_index += 1
|
69 |
+
background_image = Image.fromarray(background_frame)
|
70 |
+
processed_image = process(pil_image, background_image, fast_mode)
|
71 |
+
else:
|
72 |
+
processed_image = pil_image # Default to original image if no background is selected
|
73 |
+
return np.array(processed_image), bg_frame_index
|
74 |
+
except Exception as e:
|
75 |
+
print(f"Error processing frame: {e}")
|
76 |
+
return frame, bg_frame_index
|
77 |
|
78 |
@spaces.GPU
|
79 |
def fn(vid, bg_type="Color", bg_image=None, bg_video=None, color="#00FF00", fps=0, video_handling="slow_down", fast_mode=True):
|
|
|
91 |
audio = video.audio
|
92 |
|
93 |
# Extract frames at the specified FPS
|
94 |
+
frames = list(video.iter_frames(fps=fps))
|
95 |
|
96 |
# Process each frame for background removal
|
97 |
processed_frames = []
|
|
|
110 |
|
111 |
bg_frame_index = 0 # Initialize background frame index
|
112 |
|
113 |
+
# Use ThreadPoolExecutor for parallel processing
|
114 |
+
with ThreadPoolExecutor(max_workers=4) as executor:
|
115 |
+
futures = [executor.submit(process_frame, frames[i], bg_type, bg_image, fast_mode, bg_frame_index, background_frames, color) for i in range(len(frames))]
|
116 |
+
for future in futures:
|
117 |
+
result, bg_frame_index = future.result()
|
118 |
+
processed_frames.append(result)
|
119 |
+
elapsed_time = time.time() - start_time
|
120 |
+
yield result, None, f"Processing frame {len(processed_frames)}... Elapsed time: {elapsed_time:.2f} seconds"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
121 |
|
122 |
# Create a new video from the processed frames
|
123 |
processed_video = mp.ImageSequenceClip(processed_frames, fps=fps)
|
|
|
134 |
|
135 |
elapsed_time = time.time() - start_time
|
136 |
yield gr.update(visible=False), gr.update(visible=True), f"Processing complete! Elapsed time: {elapsed_time:.2f} seconds"
|
137 |
+
|
138 |
# Return the path to the temporary file
|
139 |
+
yield processed_frames[-1], temp_filepath, f"Processing complete! Elapsed time: {elapsed_time:.2f} seconds"
|
140 |
|
141 |
except Exception as e:
|
142 |
print(f"Error: {e}")
|
|
|
144 |
yield gr.update(visible=False), gr.update(visible=True), f"Error processing video: {e}. Elapsed time: {elapsed_time:.2f} seconds"
|
145 |
yield None, f"Error processing video: {e}", f"Error processing video: {e}. Elapsed time: {elapsed_time:.2f} seconds"
|
146 |
|
|
|
147 |
def process(image, bg, fast_mode=False):
|
148 |
image_size = image.size
|
149 |
input_images = transform_image(image).unsqueeze(0).to("cuda")
|
|
|
168 |
|
169 |
# Composite the image onto the background using the mask
|
170 |
image = Image.composite(image, background, mask)
|
|
|
171 |
return image
|
172 |
|
|
|
173 |
with gr.Blocks(theme=gr.themes.Ocean()) as demo:
|
174 |
+
gr.Markdown("# Video Background Remover & Changer\n### You can replace image background with any color, image or video.\nNOTE: As this Space is running on ZERO GPU it has limit. It can handle approx 200 frames at once. So, if you have a big video than use small chunks or Duplicate this space.")
|
175 |
with gr.Row():
|
176 |
in_video = gr.Video(label="Input Video", interactive=True)
|
177 |
stream_image = gr.Image(label="Streaming Output", visible=False)
|
|
|
193 |
with gr.Column(visible=False) as video_handling_options:
|
194 |
video_handling_radio = gr.Radio(["slow_down", "loop"], label="Video Handling", value="slow_down", interactive=True)
|
195 |
fast_mode_checkbox = gr.Checkbox(label="Fast Mode (Use BiRefNet_lite)", value=True, interactive=True)
|
196 |
+
time_textbox = gr.Textbox(label="Time Elapsed", interactive=False) # Add time textbox
|
|
|
197 |
|
198 |
def update_visibility(bg_type):
|
199 |
if bg_type == "Color":
|