Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,6 +4,7 @@ import numpy as np
|
|
| 4 |
import os
|
| 5 |
import tempfile
|
| 6 |
import subprocess
|
|
|
|
| 7 |
from basicsr.archs.rrdbnet_arch import RRDBNet
|
| 8 |
from realesrgan import RealESRGANer
|
| 9 |
from realesrgan.archs.srvgg_arch import SRVGGNetCompact
|
|
@@ -14,7 +15,7 @@ from basicsr.utils.download_util import load_file_from_url
|
|
| 14 |
# We create a dictionary to cache models so they are only loaded once.
|
| 15 |
model_cache = {}
|
| 16 |
|
| 17 |
-
def get_upsampler(model_name='realesr-general-x4v3'
|
| 18 |
"""Loads and returns the specified RealESRGAN model."""
|
| 19 |
if model_name in model_cache:
|
| 20 |
return model_cache[model_name]
|
|
@@ -32,14 +33,14 @@ def get_upsampler(model_name='realesr-general-x4v3', denoise_strength=1):
|
|
| 32 |
|
| 33 |
upsampler = RealESRGANer(
|
| 34 |
scale=netscale, model_path=model_path, model=model,
|
| 35 |
-
tile=64, tile_pad=10, pre_pad=10, half=
|
| 36 |
)
|
| 37 |
model_cache[model_name] = upsampler
|
| 38 |
return upsampler
|
| 39 |
|
| 40 |
def get_face_enhancer(upsampler, outscale):
|
| 41 |
"""Loads and returns the GFPGAN face enhancer."""
|
| 42 |
-
key = '
|
| 43 |
if key in model_cache:
|
| 44 |
return model_cache[key]
|
| 45 |
|
|
@@ -51,23 +52,23 @@ def get_face_enhancer(upsampler, outscale):
|
|
| 51 |
return face_enhancer
|
| 52 |
|
| 53 |
# --- Core Video Processing Function ---
|
| 54 |
-
def enhance_video(video_path, model_name,
|
| 55 |
"""Enhances a video frame by frame and provides progress updates."""
|
| 56 |
if not video_path:
|
| 57 |
raise gr.Error("Please upload a video to enhance.")
|
| 58 |
|
| 59 |
try:
|
| 60 |
-
upsampler = get_upsampler(model_name
|
| 61 |
|
| 62 |
face_enhancer = None
|
| 63 |
if face_enhance:
|
| 64 |
face_enhancer = get_face_enhancer(upsampler, outscale)
|
| 65 |
|
| 66 |
cap = cv2.VideoCapture(video_path)
|
| 67 |
-
fps = cap.get(
|
| 68 |
-
width = int(cap.get(
|
| 69 |
-
height = int(cap.get(
|
| 70 |
-
total_frames = int(cap.get(
|
| 71 |
|
| 72 |
# Prepare output video writer
|
| 73 |
temp_dir = tempfile.mkdtemp()
|
|
@@ -76,7 +77,7 @@ def enhance_video(video_path, model_name, denoise_strength, outscale, face_enhan
|
|
| 76 |
writer = cv2.VideoWriter(enhanced_video_path, fourcc, fps, (width * outscale, height * outscale))
|
| 77 |
|
| 78 |
# Process each frame
|
| 79 |
-
for
|
| 80 |
ret, frame = cap.read()
|
| 81 |
if not ret:
|
| 82 |
break
|
|
@@ -93,7 +94,7 @@ def enhance_video(video_path, model_name, denoise_strength, outscale, face_enhan
|
|
| 93 |
|
| 94 |
# Merge audio back into the enhanced video
|
| 95 |
final_output_path = os.path.join(temp_dir, "final_output_with_audio.mp4")
|
| 96 |
-
audio_merge_cmd = f
|
| 97 |
subprocess.call(audio_merge_cmd, shell=True, stderr=subprocess.DEVNULL, stdout=subprocess.DEVNULL)
|
| 98 |
|
| 99 |
return final_output_path
|
|
@@ -106,46 +107,54 @@ def enhance_video(video_path, model_name, denoise_strength, outscale, face_enhan
|
|
| 106 |
with gr.Blocks(theme=gr.themes.Soft(primary_hue="violet"), title="π₯ AI Video Enhancer") as demo:
|
| 107 |
gr.Markdown(
|
| 108 |
"""
|
|
|
|
| 109 |
Improve video quality, upscale resolution, and restore faces with cutting-edge AI.
|
| 110 |
**Note:** Processing can be slow, especially for longer videos.
|
| 111 |
"""
|
| 112 |
)
|
| 113 |
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
|
|
|
|
|
|
|
|
|
| 121 |
model_name = gr.Dropdown(
|
| 122 |
choices=["realesr-general-x4v3", "RealESRGAN_x4plus_anime_6B"],
|
| 123 |
value="realesr-general-x4v3",
|
| 124 |
-
label="Model Type"
|
| 125 |
)
|
| 126 |
outscale = gr.Slider(1, 4, value=2, step=1, label="Upscale Factor")
|
| 127 |
-
face_enhance = gr.Checkbox(label="Restore Faces (GFPGAN)")
|
| 128 |
-
denoise_strength = gr.Slider(0, 1, value=0.5, step=0.1, label="Denoise Strength (for general model only)")
|
| 129 |
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 136 |
|
| 137 |
# --- Event Logic ---
|
| 138 |
-
def on_submit(video, model,
|
| 139 |
# When the button is clicked, start the enhancement and return the path to the output video.
|
| 140 |
# Also, make the download button visible.
|
| 141 |
-
output_path = enhance_video(video, model,
|
| 142 |
return output_path, gr.update(value=output_path, visible=True)
|
| 143 |
|
| 144 |
enhance_btn.click(
|
| 145 |
fn=on_submit,
|
| 146 |
-
inputs=[video_input, model_name,
|
| 147 |
outputs=[video_output, download_file]
|
| 148 |
)
|
| 149 |
|
| 150 |
if __name__ == '__main__':
|
| 151 |
-
demo.launch()
|
|
|
|
| 4 |
import os
|
| 5 |
import tempfile
|
| 6 |
import subprocess
|
| 7 |
+
import traceback
|
| 8 |
from basicsr.archs.rrdbnet_arch import RRDBNet
|
| 9 |
from realesrgan import RealESRGANer
|
| 10 |
from realesrgan.archs.srvgg_arch import SRVGGNetCompact
|
|
|
|
| 15 |
# We create a dictionary to cache models so they are only loaded once.
|
| 16 |
model_cache = {}
|
| 17 |
|
| 18 |
+
def get_upsampler(model_name='realesr-general-x4v3'):
|
| 19 |
"""Loads and returns the specified RealESRGAN model."""
|
| 20 |
if model_name in model_cache:
|
| 21 |
return model_cache[model_name]
|
|
|
|
| 33 |
|
| 34 |
upsampler = RealESRGANer(
|
| 35 |
scale=netscale, model_path=model_path, model=model,
|
| 36 |
+
tile=64, tile_pad=10, pre_pad=10, half=True, gpu_id=None # Use half precision for speed
|
| 37 |
)
|
| 38 |
model_cache[model_name] = upsampler
|
| 39 |
return upsampler
|
| 40 |
|
| 41 |
def get_face_enhancer(upsampler, outscale):
|
| 42 |
"""Loads and returns the GFPGAN face enhancer."""
|
| 43 |
+
key = f'face_enhancer_{outscale}'
|
| 44 |
if key in model_cache:
|
| 45 |
return model_cache[key]
|
| 46 |
|
|
|
|
| 52 |
return face_enhancer
|
| 53 |
|
| 54 |
# --- Core Video Processing Function ---
|
| 55 |
+
def enhance_video(video_path, model_name, outscale, face_enhance, progress=gr.Progress(track_tqdm=True)):
|
| 56 |
"""Enhances a video frame by frame and provides progress updates."""
|
| 57 |
if not video_path:
|
| 58 |
raise gr.Error("Please upload a video to enhance.")
|
| 59 |
|
| 60 |
try:
|
| 61 |
+
upsampler = get_upsampler(model_name)
|
| 62 |
|
| 63 |
face_enhancer = None
|
| 64 |
if face_enhance:
|
| 65 |
face_enhancer = get_face_enhancer(upsampler, outscale)
|
| 66 |
|
| 67 |
cap = cv2.VideoCapture(video_path)
|
| 68 |
+
fps = cap.get(cv.CAP_PROP_FPS)
|
| 69 |
+
width = int(cap.get(cv.CAP_PROP_FRAME_WIDTH))
|
| 70 |
+
height = int(cap.get(cv.CAP_PROP_FRAME_HEIGHT))
|
| 71 |
+
total_frames = int(cap.get(cv.CAP_PROP_FRAME_COUNT))
|
| 72 |
|
| 73 |
# Prepare output video writer
|
| 74 |
temp_dir = tempfile.mkdtemp()
|
|
|
|
| 77 |
writer = cv2.VideoWriter(enhanced_video_path, fourcc, fps, (width * outscale, height * outscale))
|
| 78 |
|
| 79 |
# Process each frame
|
| 80 |
+
for _ in progress.tqdm(range(total_frames), desc="Enhancing Frames..."):
|
| 81 |
ret, frame = cap.read()
|
| 82 |
if not ret:
|
| 83 |
break
|
|
|
|
| 94 |
|
| 95 |
# Merge audio back into the enhanced video
|
| 96 |
final_output_path = os.path.join(temp_dir, "final_output_with_audio.mp4")
|
| 97 |
+
audio_merge_cmd = f'ffmpeg -y -i "{enhanced_video_path}" -i "{video_path}" -c:v libx264 -crf 23 -preset fast -c:a aac -b:a 128k -map 0:v:0 -map 1:a:0 -shortest "{final_output_path}"'
|
| 98 |
subprocess.call(audio_merge_cmd, shell=True, stderr=subprocess.DEVNULL, stdout=subprocess.DEVNULL)
|
| 99 |
|
| 100 |
return final_output_path
|
|
|
|
| 107 |
with gr.Blocks(theme=gr.themes.Soft(primary_hue="violet"), title="π₯ AI Video Enhancer") as demo:
|
| 108 |
gr.Markdown(
|
| 109 |
"""
|
| 110 |
+
# π₯ AI Video Enhancer & Upscaler
|
| 111 |
Improve video quality, upscale resolution, and restore faces with cutting-edge AI.
|
| 112 |
**Note:** Processing can be slow, especially for longer videos.
|
| 113 |
"""
|
| 114 |
)
|
| 115 |
|
| 116 |
+
# Top row for video previews
|
| 117 |
+
with gr.Row():
|
| 118 |
+
video_input = gr.Video(label="π¬ Original Video")
|
| 119 |
+
video_output = gr.Video(label="π Enhanced Result")
|
| 120 |
+
|
| 121 |
+
# Panel for all settings and actions
|
| 122 |
+
with gr.Box():
|
| 123 |
+
with gr.Row():
|
| 124 |
+
# Left side for main settings
|
| 125 |
+
with gr.Column(scale=3):
|
| 126 |
model_name = gr.Dropdown(
|
| 127 |
choices=["realesr-general-x4v3", "RealESRGAN_x4plus_anime_6B"],
|
| 128 |
value="realesr-general-x4v3",
|
| 129 |
+
label="Model Type (General or Anime)"
|
| 130 |
)
|
| 131 |
outscale = gr.Slider(1, 4, value=2, step=1, label="Upscale Factor")
|
|
|
|
|
|
|
| 132 |
|
| 133 |
+
# Right side for the most important actions
|
| 134 |
+
with gr.Column(scale=1, min_width=200):
|
| 135 |
+
face_enhance = gr.Checkbox(label="β¨ Restore Faces (GFPGAN)", value=False)
|
| 136 |
+
enhance_btn = gr.Button("π Enhance Video", variant="primary")
|
| 137 |
+
|
| 138 |
+
# Examples and Download components
|
| 139 |
+
gr.Examples(
|
| 140 |
+
examples=["sample_video.mp4"], # Add path to your example video
|
| 141 |
+
inputs=[video_input],
|
| 142 |
+
label="Click an example to start"
|
| 143 |
+
)
|
| 144 |
+
download_file = gr.File(label="β¬οΈ Download Enhanced Video", visible=False)
|
| 145 |
|
| 146 |
# --- Event Logic ---
|
| 147 |
+
def on_submit(video, model, scale, face):
|
| 148 |
# When the button is clicked, start the enhancement and return the path to the output video.
|
| 149 |
# Also, make the download button visible.
|
| 150 |
+
output_path = enhance_video(video, model, scale, face)
|
| 151 |
return output_path, gr.update(value=output_path, visible=True)
|
| 152 |
|
| 153 |
enhance_btn.click(
|
| 154 |
fn=on_submit,
|
| 155 |
+
inputs=[video_input, model_name, outscale, face_enhance],
|
| 156 |
outputs=[video_output, download_file]
|
| 157 |
)
|
| 158 |
|
| 159 |
if __name__ == '__main__':
|
| 160 |
+
demo.launch()
|