Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,195 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import spaces
|
2 |
+
import os
|
3 |
+
import subprocess
|
4 |
+
import tempfile
|
5 |
+
import uuid
|
6 |
+
import glob
|
7 |
+
import shutil
|
8 |
+
import time
|
9 |
+
|
10 |
+
import gradio as gr
|
11 |
+
|
12 |
+
# Set environment variables
|
13 |
+
os.environ["PIXEL3DMM_CODE_BASE"] = "./"
|
14 |
+
os.environ["PIXEL3DMM_PREPROCESSED_DATA"] = "./proprocess_results"
|
15 |
+
os.environ["PIXEL3DMM_TRACKING_OUTPUT"] = "./tracking_results"
|
16 |
+
|
17 |
+
# Utility to stitch frames into a video
|
18 |
+
def make_video_from_frames(frames_dir, out_path, fps=15):
|
19 |
+
if not os.path.isdir(frames_dir):
|
20 |
+
return None
|
21 |
+
files = glob.glob(os.path.join(frames_dir, "*.jpg")) + glob.glob(os.path.join(frames_dir, "*.png"))
|
22 |
+
if not files:
|
23 |
+
return None
|
24 |
+
ext = files[0].split('.')[-1]
|
25 |
+
pattern = os.path.join(frames_dir, f"%05d.{ext}")
|
26 |
+
subprocess.run([
|
27 |
+
"ffmpeg", "-y", "-i", pattern,
|
28 |
+
"-r", str(fps), out_path
|
29 |
+
], check=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
|
30 |
+
return out_path
|
31 |
+
|
32 |
+
# Function to probe video for duration and frame rate
|
33 |
+
def get_video_info(video_path):
|
34 |
+
"""
|
35 |
+
Probes the uploaded video and returns updated slider configs:
|
36 |
+
- seconds slider: max = int(duration)
|
37 |
+
- fps slider: max = int(orig_fps)
|
38 |
+
"""
|
39 |
+
if not video_path:
|
40 |
+
# Return default slider updates when no video is uploaded
|
41 |
+
return gr.update(maximum=10, value=3, step=1), gr.update(maximum=30, value=15, step=1)
|
42 |
+
|
43 |
+
# Use ffprobe to get JSON metadata
|
44 |
+
cmd = [
|
45 |
+
"ffprobe", "-v", "quiet",
|
46 |
+
"-print_format", "json",
|
47 |
+
"-show_streams", video_path
|
48 |
+
]
|
49 |
+
res = subprocess.run(cmd, capture_output=True, text=True)
|
50 |
+
try:
|
51 |
+
import json
|
52 |
+
data = json.loads(res.stdout)
|
53 |
+
stream = next(s for s in data.get('streams', []) if s.get('codec_type') == 'video')
|
54 |
+
duration = float(stream.get('duration') or data.get('format', {}).get('duration', 0))
|
55 |
+
fr = stream.get('r_frame_rate', '0/1')
|
56 |
+
num, den = fr.split('/')
|
57 |
+
orig_fps = float(num) / float(den) if float(den) else 30
|
58 |
+
except Exception:
|
59 |
+
duration, orig_fps = 10, 30
|
60 |
+
|
61 |
+
# Configure sliders based on actual video properties
|
62 |
+
seconds_cfg = gr.update(maximum=int(duration), value=min(int(duration), 3), step=1)
|
63 |
+
fps_cfg = gr.update(maximum=int(orig_fps), value=min(int(orig_fps), 15), step=1)
|
64 |
+
return seconds_cfg, fps_cfg
|
65 |
+
|
66 |
+
# Step 1: Trim video based on user-defined duration and fps based on user-defined duration and fps
|
67 |
+
@space.GPU()
|
68 |
+
def step1_trim(video_path, seconds, fps, state):
|
69 |
+
session_id = str(uuid.uuid4())
|
70 |
+
base_dir = os.path.join(os.environ["PIXEL3DMM_PREPROCESSED_DATA"], session_id)
|
71 |
+
state.update({"session_id": session_id, "base_dir": base_dir})
|
72 |
+
|
73 |
+
tmp = tempfile.mkdtemp()
|
74 |
+
trimmed = os.path.join(tmp, f"{session_id}.mp4")
|
75 |
+
subprocess.run([
|
76 |
+
"ffmpeg", "-y", "-i", video_path,
|
77 |
+
"-t", str(seconds), # user-specified duration
|
78 |
+
"-r", str(fps), # user-specified fps
|
79 |
+
trimmed
|
80 |
+
], check=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
|
81 |
+
state["trimmed_path"] = trimmed
|
82 |
+
return f"β
Step 1: Trimmed to {seconds}s @{fps}fps", state
|
83 |
+
|
84 |
+
# Step 2: Preprocessing β cropped video
|
85 |
+
@space.GPU()
|
86 |
+
def step2_preprocess(state):
|
87 |
+
session_id = state["session_id"]
|
88 |
+
base_dir = state["base_dir"]
|
89 |
+
trimmed = state["trimmed_path"]
|
90 |
+
|
91 |
+
subprocess.run([
|
92 |
+
"python", "scripts/run_preprocessing.py",
|
93 |
+
"--video_or_images_path", trimmed
|
94 |
+
], check=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
|
95 |
+
|
96 |
+
crop_dir = os.path.join(base_dir, "cropped")
|
97 |
+
out = os.path.join(os.path.dirname(trimmed), f"crop_{session_id}.mp4")
|
98 |
+
video = make_video_from_frames(crop_dir, out)
|
99 |
+
return "β
Step 2: Preprocessing complete", video, state
|
100 |
+
|
101 |
+
# Step 3: Normals inference β normals video
|
102 |
+
@space.GPU()
|
103 |
+
def step3_normals(state):
|
104 |
+
session_id = state["session_id"]
|
105 |
+
base_dir = state["base_dir"]
|
106 |
+
|
107 |
+
subprocess.run([
|
108 |
+
"python", "scripts/network_inference.py",
|
109 |
+
"model.prediction_type=normals", f"video_name={session_id}"
|
110 |
+
], check=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
|
111 |
+
|
112 |
+
normals_dir = os.path.join(base_dir, "p3dmm", "normals")
|
113 |
+
out = os.path.join(os.path.dirname(state["trimmed_path"]), f"normals_{session_id}.mp4")
|
114 |
+
video = make_video_from_frames(normals_dir, out)
|
115 |
+
return "β
Step 3: Normals inference complete", video, state
|
116 |
+
|
117 |
+
# Step 4: UV map inference β uv map video
|
118 |
+
@space.GPU()
|
119 |
+
def step4_uv_map(state):
|
120 |
+
session_id = state["session_id"]
|
121 |
+
base_dir = state["base_dir"]
|
122 |
+
|
123 |
+
subprocess.run([
|
124 |
+
"python", "scripts/network_inference.py",
|
125 |
+
"model.prediction_type=uv_map", f"video_name={session_id}"
|
126 |
+
], check=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
|
127 |
+
|
128 |
+
uv_dir = os.path.join(base_dir, "p3dmm", "uv_map")
|
129 |
+
out = os.path.join(os.path.dirname(state["trimmed_path"]), f"uv_map_{session_id}.mp4")
|
130 |
+
video = make_video_from_frames(uv_dir, out)
|
131 |
+
return "β
Step 4: UV map inference complete", video, state
|
132 |
+
|
133 |
+
# Step 5: Tracking β final tracking video
|
134 |
+
@space.GPU()
|
135 |
+
def step5_track(state):
|
136 |
+
session_id = state["session_id"]
|
137 |
+
script = os.path.join(os.environ["PIXEL3DMM_CODE_BASE"], "scripts", "track.py")
|
138 |
+
cmd = [
|
139 |
+
"python", script,
|
140 |
+
f"video_name={session_id}"
|
141 |
+
]
|
142 |
+
try:
|
143 |
+
# capture both stdout & stderr
|
144 |
+
p = subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True, check=True)
|
145 |
+
except subprocess.CalledProcessError as e:
|
146 |
+
# e.stdout contains everything
|
147 |
+
err = f"β Tracking failed (exit {e.returncode}).\n\n{e.stdout}"
|
148 |
+
return err, None, state
|
149 |
+
|
150 |
+
# if we get here, it succeeded:
|
151 |
+
tracking_dir = os.path.join(os.environ["PIXEL3DMM_TRACKING_OUTPUT"], session_id, "frames")
|
152 |
+
out = os.path.join(os.path.dirname(state["trimmed_path"]), f"result_{session_id}.mp4")
|
153 |
+
video = make_video_from_frames(tracking_dir, out)
|
154 |
+
return "β
Step 5: Tracking complete", video, state
|
155 |
+
|
156 |
+
# Build Gradio UI
|
157 |
+
demo = gr.Blocks()
|
158 |
+
|
159 |
+
with demo:
|
160 |
+
gr.Markdown("## Video Processing Pipeline")
|
161 |
+
with gr.Row():
|
162 |
+
with gr.Column():
|
163 |
+
video_in = gr.Video(label="Upload video", height=512)
|
164 |
+
# Sliders for duration and fps
|
165 |
+
seconds_slider = gr.Slider(label="Duration (seconds)", minimum=2, maximum=10, step=1, value=3)
|
166 |
+
fps_slider = gr.Slider(label="Frame Rate (fps)", minimum=15, maximum=30, step=1, value=15)
|
167 |
+
status = gr.Textbox(label="Status", lines=2, interactive=False)
|
168 |
+
state = gr.State({})
|
169 |
+
with gr.Column():
|
170 |
+
with gr.Row():
|
171 |
+
crop_vid = gr.Video(label="Preprocessed", height=256)
|
172 |
+
normals_vid = gr.Video(label="Normals", height=256)
|
173 |
+
with gr.Row():
|
174 |
+
uv_vid = gr.Video(label="UV Map", height=256)
|
175 |
+
track_vid = gr.Video(label="Tracking", height=256)
|
176 |
+
run_btn = gr.Button("Run Pipeline")
|
177 |
+
|
178 |
+
# Update sliders after video upload
|
179 |
+
video_in.change(fn=get_video_info, inputs=video_in, outputs=[seconds_slider, fps_slider])
|
180 |
+
|
181 |
+
# Pipeline execution
|
182 |
+
(run_btn.click(fn=step1_trim, inputs=[video_in, seconds_slider, fps_slider, state], outputs=[status, state])
|
183 |
+
.then(fn=step2_preprocess, inputs=[state], outputs=[status, crop_vid, state])
|
184 |
+
.then(fn=step3_normals, inputs=[state], outputs=[status, normals_vid, state])
|
185 |
+
.then(fn=step4_uv_map, inputs=[state], outputs=[status, uv_vid, state])
|
186 |
+
.then(fn=step5_track, inputs=[state], outputs=[status, track_vid, state])
|
187 |
+
)
|
188 |
+
|
189 |
+
# ------------------------------------------------------------------
|
190 |
+
# START THE GRADIO SERVER
|
191 |
+
# ------------------------------------------------------------------
|
192 |
+
demo.queue()
|
193 |
+
|
194 |
+
demo.launch(share=True)
|
195 |
+
|