Spaces:
Running
on
Zero
Running
on
Zero
Update gradio_app.py
Browse files- gradio_app.py +69 -77
gradio_app.py
CHANGED
@@ -1,16 +1,15 @@
|
|
1 |
import spaces
|
2 |
import torch
|
3 |
-
import os
|
4 |
import time
|
5 |
-
import
|
6 |
-
from
|
7 |
-
from propainter.inference import Propainter, get_device
|
8 |
import gradio as gr
|
9 |
|
10 |
# Download Weights
|
11 |
from huggingface_hub import snapshot_download
|
12 |
|
13 |
-
# List of subdirectories to create inside "
|
14 |
subfolders = [
|
15 |
"diffuEraser",
|
16 |
"stable-diffusion-v1-5",
|
@@ -18,46 +17,35 @@ subfolders = [
|
|
18 |
"propainter",
|
19 |
"sd-vae-ft-mse"
|
20 |
]
|
21 |
-
# Create
|
22 |
for subfolder in subfolders:
|
23 |
-
os.makedirs(os.path.join("
|
24 |
-
|
25 |
-
snapshot_download(
|
26 |
-
repo_id = "lixiaowen/diffuEraser",
|
27 |
-
local_dir = "./weights/diffuEraser"
|
28 |
-
)
|
29 |
-
|
30 |
-
snapshot_download(
|
31 |
-
repo_id = "stable-diffusion-v1-5/stable-diffusion-v1-5",
|
32 |
-
local_dir = "./weights/stable-diffusion-v1-5"
|
33 |
-
)
|
34 |
|
35 |
-
snapshot_download(
|
36 |
-
|
37 |
-
|
38 |
-
)
|
|
|
39 |
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
)
|
44 |
-
|
45 |
-
snapshot_download(
|
46 |
-
repo_id = "stabilityai/sd-vae-ft-mse",
|
47 |
-
local_dir = "./weights/sd-vae-ft-mse"
|
48 |
-
)
|
49 |
|
50 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
|
52 |
@spaces.GPU(duration=120)
|
53 |
def infer(input_video, input_mask):
|
54 |
-
|
55 |
-
|
56 |
-
mask_dilation_iter = 8
|
57 |
-
max_img_size = 960
|
58 |
-
|
59 |
-
|
60 |
-
ref_stride = 10
|
61 |
neighbor_length = 10
|
62 |
subvideo_length = 50
|
63 |
|
@@ -65,50 +53,66 @@ def infer(input_video, input_mask):
|
|
65 |
vae_path = "weights/sd-vae-ft-mse"
|
66 |
diffueraser_path = "weights/diffuEraser"
|
67 |
propainter_model_dir = "weights/propainter"
|
68 |
-
|
69 |
if not os.path.exists(save_path):
|
70 |
os.makedirs(save_path)
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
device = get_device()
|
76 |
-
# PCM params
|
77 |
ckpt = "2-Step"
|
78 |
video_inpainting_sd = DiffuEraser(device, base_model_path, vae_path, diffueraser_path, ckpt=ckpt)
|
79 |
propainter = Propainter(propainter_model_dir, device=device)
|
80 |
-
|
81 |
-
start_time = time.time()
|
82 |
|
83 |
-
|
84 |
-
|
85 |
-
ref_stride=ref_stride, neighbor_length=neighbor_length, subvideo_length = subvideo_length,
|
86 |
-
mask_dilation = mask_dilation_iter)
|
87 |
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
guidance_scale=guidance_scale)
|
93 |
-
|
94 |
-
end_time = time.time()
|
95 |
-
inference_time = end_time - start_time
|
96 |
-
print(f"DiffuEraser inference time: {inference_time:.4f} s")
|
97 |
|
98 |
-
|
|
|
99 |
|
|
|
100 |
return output_path
|
101 |
|
|
|
102 |
with gr.Blocks() as demo:
|
103 |
-
|
104 |
with gr.Column():
|
105 |
gr.Markdown("# DiffuEraser: A Diffusion Model for Video Inpainting")
|
106 |
-
gr.Markdown("DiffuEraser is a diffusion model for video inpainting, which outperforms state-of-the-art model
|
|
|
107 |
gr.HTML("""
|
108 |
<div style="display:flex;column-gap:4px;">
|
109 |
<a href="https://github.com/lixiaowen-xw/DiffuEraser">
|
110 |
<img src='https://img.shields.io/badge/GitHub-Repo-blue'>
|
111 |
-
</a>
|
112 |
<a href="https://lixiaowen-xw.github.io/DiffuEraser-page">
|
113 |
<img src='https://img.shields.io/badge/Project-Page-green'>
|
114 |
</a>
|
@@ -122,34 +126,22 @@ with gr.Blocks() as demo:
|
|
122 |
""")
|
123 |
|
124 |
with gr.Row():
|
125 |
-
|
126 |
with gr.Column():
|
127 |
-
|
128 |
input_video = gr.Video(label="Input Video (MP4 ONLY)")
|
129 |
input_mask = gr.Video(label="Input Mask Video (MP4 ONLY)")
|
130 |
submit_btn = gr.Button("Submit")
|
131 |
|
132 |
with gr.Column():
|
133 |
-
|
134 |
video_result = gr.Video(label="Result")
|
135 |
gr.Examples(
|
136 |
-
examples
|
137 |
["./examples/example1/video.mp4", "./examples/example1/mask.mp4"],
|
138 |
["./examples/example2/video.mp4", "./examples/example2/mask.mp4"],
|
139 |
["./examples/example3/video.mp4", "./examples/example3/mask.mp4"],
|
140 |
],
|
141 |
-
inputs
|
142 |
)
|
143 |
|
144 |
-
|
145 |
-
submit_btn.click(
|
146 |
-
fn = infer,
|
147 |
-
inputs = [input_video, input_mask],
|
148 |
-
outputs = [video_result]
|
149 |
-
)
|
150 |
|
151 |
demo.queue().launch(show_api=False, show_error=True)
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
|
|
1 |
import spaces
|
2 |
import torch
|
3 |
+
import os
|
4 |
import time
|
5 |
+
import datetime
|
6 |
+
from moviepy.editor import VideoFileClip
|
|
|
7 |
import gradio as gr
|
8 |
|
9 |
# Download Weights
|
10 |
from huggingface_hub import snapshot_download
|
11 |
|
12 |
+
# List of subdirectories to create inside "weights"
|
13 |
subfolders = [
|
14 |
"diffuEraser",
|
15 |
"stable-diffusion-v1-5",
|
|
|
17 |
"propainter",
|
18 |
"sd-vae-ft-mse"
|
19 |
]
|
20 |
+
# Create directories
|
21 |
for subfolder in subfolders:
|
22 |
+
os.makedirs(os.path.join("weights", subfolder), exist_ok=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
|
24 |
+
snapshot_download(repo_id="lixiaowen/diffuEraser", local_dir="./weights/diffuEraser")
|
25 |
+
snapshot_download(repo_id="stable-diffusion-v1-5/stable-diffusion-v1-5", local_dir="./weights/stable-diffusion-v1-5")
|
26 |
+
snapshot_download(repo_id="wangfuyun/PCM_Weights", local_dir="./weights/PCM_Weights")
|
27 |
+
snapshot_download(repo_id="camenduru/ProPainter", local_dir="./weights/propainter")
|
28 |
+
snapshot_download(repo_id="stabilityai/sd-vae-ft-mse", local_dir="./weights/sd-vae-ft-mse")
|
29 |
|
30 |
+
# Import model classes
|
31 |
+
from diffueraser.diffueraser import DiffuEraser
|
32 |
+
from propainter.inference import Propainter, get_device
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
+
# Helper function to trim videos
|
35 |
+
def trim_video(input_path, output_path, max_duration=5):
|
36 |
+
clip = VideoFileClip(input_path)
|
37 |
+
trimmed_clip = clip.subclip(0, min(max_duration, clip.duration))
|
38 |
+
trimmed_clip.write_videofile(output_path, codec="libx264", audio_codec="aac")
|
39 |
+
clip.close()
|
40 |
+
trimmed_clip.close()
|
41 |
|
42 |
@spaces.GPU(duration=120)
|
43 |
def infer(input_video, input_mask):
|
44 |
+
# Setup paths and parameters
|
45 |
+
save_path = "results"
|
46 |
+
mask_dilation_iter = 8
|
47 |
+
max_img_size = 960
|
48 |
+
ref_stride = 10
|
|
|
|
|
49 |
neighbor_length = 10
|
50 |
subvideo_length = 50
|
51 |
|
|
|
53 |
vae_path = "weights/sd-vae-ft-mse"
|
54 |
diffueraser_path = "weights/diffuEraser"
|
55 |
propainter_model_dir = "weights/propainter"
|
56 |
+
|
57 |
if not os.path.exists(save_path):
|
58 |
os.makedirs(save_path)
|
59 |
+
|
60 |
+
# Timestamp for unique filenames
|
61 |
+
timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
|
62 |
+
trimmed_video_path = os.path.join(save_path, f"trimmed_video_{timestamp}.mp4")
|
63 |
+
trimmed_mask_path = os.path.join(save_path, f"trimmed_mask_{timestamp}.mp4")
|
64 |
+
priori_path = os.path.join(save_path, f"priori_{timestamp}.mp4")
|
65 |
+
output_path = os.path.join(save_path, f"diffueraser_result_{timestamp}.mp4")
|
66 |
+
|
67 |
+
# Trim input videos
|
68 |
+
trim_video(input_video, trimmed_video_path)
|
69 |
+
trim_video(input_mask, trimmed_mask_path)
|
70 |
+
|
71 |
+
# Dynamically compute video_length (in frames) assuming 30 fps
|
72 |
+
clip = VideoFileClip(trimmed_video_path)
|
73 |
+
video_duration = clip.duration
|
74 |
+
clip.close()
|
75 |
+
video_length = int(video_duration * 30)
|
76 |
+
|
77 |
+
# Model setup
|
78 |
device = get_device()
|
|
|
79 |
ckpt = "2-Step"
|
80 |
video_inpainting_sd = DiffuEraser(device, base_model_path, vae_path, diffueraser_path, ckpt=ckpt)
|
81 |
propainter = Propainter(propainter_model_dir, device=device)
|
|
|
|
|
82 |
|
83 |
+
# Run models
|
84 |
+
start_time = time.time()
|
|
|
|
|
85 |
|
86 |
+
# ProPainter (priori)
|
87 |
+
propainter.forward(trimmed_video_path, trimmed_mask_path, priori_path,
|
88 |
+
video_length=video_length, ref_stride=ref_stride,
|
89 |
+
neighbor_length=neighbor_length, subvideo_length=subvideo_length,
|
90 |
+
mask_dilation=mask_dilation_iter)
|
91 |
+
|
92 |
+
# DiffuEraser
|
93 |
+
guidance_scale = None
|
94 |
+
video_inpainting_sd.forward(trimmed_video_path, trimmed_mask_path, priori_path, output_path,
|
95 |
+
max_img_size=max_img_size, video_length=video_length,
|
96 |
+
mask_dilation_iter=mask_dilation_iter,
|
97 |
guidance_scale=guidance_scale)
|
|
|
|
|
|
|
|
|
98 |
|
99 |
+
end_time = time.time()
|
100 |
+
print(f"DiffuEraser inference time: {end_time - start_time:.2f} seconds")
|
101 |
|
102 |
+
torch.cuda.empty_cache()
|
103 |
return output_path
|
104 |
|
105 |
+
# Gradio interface
|
106 |
with gr.Blocks() as demo:
|
|
|
107 |
with gr.Column():
|
108 |
gr.Markdown("# DiffuEraser: A Diffusion Model for Video Inpainting")
|
109 |
+
gr.Markdown("DiffuEraser is a diffusion model for video inpainting, which outperforms state-of-the-art model ProPainter in both content completeness and temporal consistency while maintaining acceptable efficiency.")
|
110 |
+
|
111 |
gr.HTML("""
|
112 |
<div style="display:flex;column-gap:4px;">
|
113 |
<a href="https://github.com/lixiaowen-xw/DiffuEraser">
|
114 |
<img src='https://img.shields.io/badge/GitHub-Repo-blue'>
|
115 |
+
</a>
|
116 |
<a href="https://lixiaowen-xw.github.io/DiffuEraser-page">
|
117 |
<img src='https://img.shields.io/badge/Project-Page-green'>
|
118 |
</a>
|
|
|
126 |
""")
|
127 |
|
128 |
with gr.Row():
|
|
|
129 |
with gr.Column():
|
|
|
130 |
input_video = gr.Video(label="Input Video (MP4 ONLY)")
|
131 |
input_mask = gr.Video(label="Input Mask Video (MP4 ONLY)")
|
132 |
submit_btn = gr.Button("Submit")
|
133 |
|
134 |
with gr.Column():
|
|
|
135 |
video_result = gr.Video(label="Result")
|
136 |
gr.Examples(
|
137 |
+
examples=[
|
138 |
["./examples/example1/video.mp4", "./examples/example1/mask.mp4"],
|
139 |
["./examples/example2/video.mp4", "./examples/example2/mask.mp4"],
|
140 |
["./examples/example3/video.mp4", "./examples/example3/mask.mp4"],
|
141 |
],
|
142 |
+
inputs=[input_video, input_mask]
|
143 |
)
|
144 |
|
145 |
+
submit_btn.click(fn=infer, inputs=[input_video, input_mask], outputs=[video_result])
|
|
|
|
|
|
|
|
|
|
|
146 |
|
147 |
demo.queue().launch(show_api=False, show_error=True)
|
|
|
|
|
|
|
|