Spaces:
Running
on
Zero
Running
on
Zero
Beijia11
commited on
Commit
·
c4fce07
1
Parent(s):
6ded12b
merge demo.py and app.py
Browse files
app.py
CHANGED
@@ -2,20 +2,30 @@ import os
|
|
2 |
import sys
|
3 |
import gradio as gr
|
4 |
import torch
|
5 |
-
import subprocess
|
6 |
import argparse
|
7 |
-
import
|
8 |
-
import
|
|
|
|
|
|
|
9 |
|
10 |
project_root = os.path.dirname(os.path.abspath(__file__))
|
11 |
os.environ["GRADIO_TEMP_DIR"] = os.path.join(project_root, "tmp", "gradio")
|
12 |
sys.path.append(project_root)
|
13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
HERE_PATH = os.path.normpath(os.path.dirname(__file__))
|
15 |
sys.path.insert(0, HERE_PATH)
|
16 |
from huggingface_hub import hf_hub_download
|
17 |
hf_hub_download(repo_id="EXCAI/Diffusion-As-Shader", filename='spatracker/spaT_final.pth', local_dir=f'{HERE_PATH}/checkpoints/')
|
18 |
|
|
|
|
|
19 |
|
20 |
# Parse command line arguments
|
21 |
parser = argparse.ArgumentParser(description="Diffusion as Shader Web UI")
|
@@ -31,21 +41,53 @@ GPU_ID = args.gpu
|
|
31 |
DEFAULT_MODEL_PATH = args.model_path
|
32 |
OUTPUT_DIR = args.output_dir
|
33 |
|
34 |
-
# if 'CUDA_HOME' not in os.environ:
|
35 |
-
# for cuda_path in ['/usr/local/cuda', '/usr/cuda', '/opt/cuda']:
|
36 |
-
# if os.path.exists(cuda_path):
|
37 |
-
# os.environ['CUDA_HOME'] = cuda_path
|
38 |
-
# print(cuda_path)
|
39 |
-
# break
|
40 |
-
# if 'CUDA_HOME' not in os.environ:
|
41 |
-
# os.environ['CUDA_HOME'] = '/usr/local/cuda'
|
42 |
-
# print("set default cuda path in: /usr/local/cuda")
|
43 |
-
|
44 |
# Create necessary directories
|
45 |
os.makedirs("outputs", exist_ok=True)
|
46 |
# Create project tmp directory instead of using system temp
|
47 |
os.makedirs(os.path.join(project_root, "tmp"), exist_ok=True)
|
48 |
os.makedirs(os.path.join(project_root, "tmp", "gradio"), exist_ok=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
49 |
|
50 |
def save_uploaded_file(file):
|
51 |
if file is None:
|
@@ -86,59 +128,22 @@ def save_uploaded_file(file):
|
|
86 |
|
87 |
return temp_path
|
88 |
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
if isinstance(value, bool):
|
105 |
-
cmd.append(f"--{key}")
|
106 |
-
cmd.append(str(value).lower()) # Convert True/False to true/false
|
107 |
-
else:
|
108 |
-
cmd.append(f"--{key}")
|
109 |
-
cmd.append(str(value))
|
110 |
-
|
111 |
-
return cmd
|
112 |
-
|
113 |
-
@spaces.GPU(duration=240)
|
114 |
-
def run_process(cmd):
|
115 |
-
"""Run command and return output"""
|
116 |
-
print(f"Running command: {' '.join(cmd)}")
|
117 |
-
process = subprocess.Popen(
|
118 |
-
cmd,
|
119 |
-
stdout=subprocess.PIPE,
|
120 |
-
stderr=subprocess.PIPE,
|
121 |
-
universal_newlines=True
|
122 |
-
)
|
123 |
-
|
124 |
-
output = []
|
125 |
-
for line in iter(process.stdout.readline, ""):
|
126 |
-
print(line, end="")
|
127 |
-
output.append(line)
|
128 |
-
if not line:
|
129 |
-
break
|
130 |
-
|
131 |
-
process.stdout.close()
|
132 |
-
return_code = process.wait()
|
133 |
-
|
134 |
-
if return_code:
|
135 |
-
stderr = process.stderr.read()
|
136 |
-
print(f"Error: {stderr}")
|
137 |
-
raise subprocess.CalledProcessError(return_code, cmd, output="\n".join(output), stderr=stderr)
|
138 |
-
|
139 |
-
return "\n".join(output)
|
140 |
|
141 |
-
@spaces.GPU(duration=240)
|
142 |
def process_motion_transfer(source, prompt, mt_repaint_option, mt_repaint_image):
|
143 |
"""Process video motion transfer task"""
|
144 |
try:
|
@@ -150,42 +155,68 @@ def process_motion_transfer(source, prompt, mt_repaint_option, mt_repaint_image)
|
|
150 |
print(f"DEBUG: Repaint option: {mt_repaint_option}")
|
151 |
print(f"DEBUG: Repaint image: {mt_repaint_image}")
|
152 |
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
"
|
158 |
-
"
|
159 |
-
|
|
|
160 |
|
161 |
-
|
162 |
if mt_repaint_image is not None:
|
163 |
-
# Custom image takes precedence if provided
|
164 |
repaint_path = save_uploaded_file(mt_repaint_image)
|
165 |
-
|
166 |
-
|
167 |
elif mt_repaint_option == "Yes":
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
181 |
else:
|
182 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
183 |
except Exception as e:
|
184 |
import traceback
|
185 |
print(f"Processing failed: {str(e)}\n{traceback.format_exc()}")
|
186 |
return None
|
187 |
|
188 |
-
@spaces.GPU(duration=240)
|
189 |
def process_camera_control(source, prompt, camera_motion, tracking_method):
|
190 |
"""Process camera control task"""
|
191 |
try:
|
@@ -197,36 +228,66 @@ def process_camera_control(source, prompt, camera_motion, tracking_method):
|
|
197 |
print(f"DEBUG: Camera motion: '{camera_motion}'")
|
198 |
print(f"DEBUG: Tracking method: '{tracking_method}'")
|
199 |
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
"
|
205 |
-
"
|
206 |
-
|
207 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
208 |
|
209 |
-
if camera_motion and camera_motion.strip():
|
210 |
-
args["camera_motion"] = camera_motion
|
211 |
|
212 |
-
|
213 |
-
|
214 |
-
|
|
|
|
|
|
|
|
|
|
|
215 |
|
216 |
-
|
217 |
-
output_files = glob.glob(os.path.join(OUTPUT_DIR, "*.mp4"))
|
218 |
-
if output_files:
|
219 |
-
# Sort by modification time, return the latest file
|
220 |
-
latest_file = max(output_files, key=os.path.getmtime)
|
221 |
-
return latest_file
|
222 |
-
else:
|
223 |
-
return None
|
224 |
except Exception as e:
|
225 |
import traceback
|
226 |
print(f"Processing failed: {str(e)}\n{traceback.format_exc()}")
|
227 |
return None
|
228 |
|
229 |
-
@spaces.GPU(duration=240)
|
230 |
def process_object_manipulation(source, prompt, object_motion, object_mask, tracking_method):
|
231 |
"""Process object manipulation task"""
|
232 |
try:
|
@@ -236,36 +297,90 @@ def process_object_manipulation(source, prompt, object_motion, object_mask, trac
|
|
236 |
return None
|
237 |
|
238 |
object_mask_path = save_uploaded_file(object_mask)
|
|
|
|
|
|
|
239 |
|
240 |
-
args = {
|
241 |
-
"input_path": input_image_path,
|
242 |
-
"prompt": prompt,
|
243 |
-
"checkpoint_path": DEFAULT_MODEL_PATH,
|
244 |
-
"output_dir": OUTPUT_DIR,
|
245 |
-
"gpu": GPU_ID,
|
246 |
-
"object_motion": object_motion,
|
247 |
-
"object_mask": object_mask_path,
|
248 |
-
"tracking_method": tracking_method
|
249 |
-
}
|
250 |
|
251 |
-
|
252 |
-
|
253 |
-
|
|
|
|
|
|
|
254 |
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
|
259 |
-
|
260 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
261 |
else:
|
262 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
263 |
except Exception as e:
|
264 |
import traceback
|
265 |
print(f"Processing failed: {str(e)}\n{traceback.format_exc()}")
|
266 |
return None
|
267 |
|
268 |
-
@spaces.GPU(duration=240)
|
269 |
def process_mesh_animation(source, prompt, tracking_video, ma_repaint_option, ma_repaint_image):
|
270 |
"""Process mesh animation task"""
|
271 |
try:
|
@@ -278,36 +393,34 @@ def process_mesh_animation(source, prompt, tracking_video, ma_repaint_option, ma
|
|
278 |
if tracking_video_path is None:
|
279 |
return None
|
280 |
|
281 |
-
args = {
|
282 |
-
"input_path": input_video_path,
|
283 |
-
"prompt": prompt,
|
284 |
-
"checkpoint_path": DEFAULT_MODEL_PATH,
|
285 |
-
"output_dir": OUTPUT_DIR,
|
286 |
-
"gpu": GPU_ID,
|
287 |
-
"tracking_path": tracking_video_path
|
288 |
-
}
|
289 |
|
290 |
-
|
|
|
|
|
|
|
291 |
if ma_repaint_image is not None:
|
292 |
-
# Custom image takes precedence if provided
|
293 |
repaint_path = save_uploaded_file(ma_repaint_image)
|
294 |
-
|
|
|
295 |
elif ma_repaint_option == "Yes":
|
296 |
-
|
297 |
-
|
298 |
-
|
299 |
-
|
300 |
-
|
301 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
302 |
|
303 |
-
|
304 |
-
output_files = glob.glob(os.path.join(OUTPUT_DIR, "*.mp4"))
|
305 |
-
if output_files:
|
306 |
-
# Sort by modification time, return the latest file
|
307 |
-
latest_file = max(output_files, key=os.path.getmtime)
|
308 |
-
return latest_file
|
309 |
-
else:
|
310 |
-
return None
|
311 |
except Exception as e:
|
312 |
import traceback
|
313 |
print(f"Processing failed: {str(e)}\n{traceback.format_exc()}")
|
|
|
2 |
import sys
|
3 |
import gradio as gr
|
4 |
import torch
|
|
|
5 |
import argparse
|
6 |
+
from PIL import Image
|
7 |
+
import numpy as np
|
8 |
+
import torchvision.transforms as transforms
|
9 |
+
from moviepy.editor import VideoFileClip
|
10 |
+
from diffusers.utils import load_image, load_video
|
11 |
|
12 |
project_root = os.path.dirname(os.path.abspath(__file__))
|
13 |
os.environ["GRADIO_TEMP_DIR"] = os.path.join(project_root, "tmp", "gradio")
|
14 |
sys.path.append(project_root)
|
15 |
|
16 |
+
try:
|
17 |
+
sys.path.append(os.path.join(project_root, "submodules/MoGe"))
|
18 |
+
os.environ["TOKENIZERS_PARALLELISM"] = "false"
|
19 |
+
except:
|
20 |
+
print("Warning: MoGe not found, motion transfer will not be applied")
|
21 |
+
|
22 |
HERE_PATH = os.path.normpath(os.path.dirname(__file__))
|
23 |
sys.path.insert(0, HERE_PATH)
|
24 |
from huggingface_hub import hf_hub_download
|
25 |
hf_hub_download(repo_id="EXCAI/Diffusion-As-Shader", filename='spatracker/spaT_final.pth', local_dir=f'{HERE_PATH}/checkpoints/')
|
26 |
|
27 |
+
from models.pipelines import DiffusionAsShaderPipeline, FirstFrameRepainter, CameraMotionGenerator, ObjectMotionGenerator
|
28 |
+
from submodules.MoGe.moge.model import MoGeModel
|
29 |
|
30 |
# Parse command line arguments
|
31 |
parser = argparse.ArgumentParser(description="Diffusion as Shader Web UI")
|
|
|
41 |
DEFAULT_MODEL_PATH = args.model_path
|
42 |
OUTPUT_DIR = args.output_dir
|
43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
# Create necessary directories
|
45 |
os.makedirs("outputs", exist_ok=True)
|
46 |
# Create project tmp directory instead of using system temp
|
47 |
os.makedirs(os.path.join(project_root, "tmp"), exist_ok=True)
|
48 |
os.makedirs(os.path.join(project_root, "tmp", "gradio"), exist_ok=True)
|
49 |
+
def load_media(media_path, max_frames=49, transform=None):
|
50 |
+
"""Load video or image frames and convert to tensor
|
51 |
+
|
52 |
+
Args:
|
53 |
+
media_path (str): Path to video or image file
|
54 |
+
max_frames (int): Maximum number of frames to load
|
55 |
+
transform (callable): Transform to apply to frames
|
56 |
+
|
57 |
+
Returns:
|
58 |
+
Tuple[torch.Tensor, float, bool]: Video tensor [T,C,H,W], FPS, and is_video flag
|
59 |
+
"""
|
60 |
+
if transform is None:
|
61 |
+
transform = transforms.Compose([
|
62 |
+
transforms.Resize((480, 720)),
|
63 |
+
transforms.ToTensor()
|
64 |
+
])
|
65 |
+
|
66 |
+
# Determine if input is video or image based on extension
|
67 |
+
ext = os.path.splitext(media_path)[1].lower()
|
68 |
+
is_video = ext in ['.mp4', '.avi', '.mov']
|
69 |
+
|
70 |
+
if is_video:
|
71 |
+
frames = load_video(media_path)
|
72 |
+
fps = len(frames) / VideoFileClip(media_path).duration
|
73 |
+
else:
|
74 |
+
# Handle image as single frame
|
75 |
+
image = load_image(media_path)
|
76 |
+
frames = [image]
|
77 |
+
fps = 8 # Default fps for images
|
78 |
+
|
79 |
+
# Ensure we have exactly max_frames
|
80 |
+
if len(frames) > max_frames:
|
81 |
+
frames = frames[:max_frames]
|
82 |
+
elif len(frames) < max_frames:
|
83 |
+
last_frame = frames[-1]
|
84 |
+
while len(frames) < max_frames:
|
85 |
+
frames.append(last_frame.copy())
|
86 |
+
|
87 |
+
# Convert frames to tensor
|
88 |
+
video_tensor = torch.stack([transform(frame) for frame in frames])
|
89 |
+
|
90 |
+
return video_tensor, fps, is_video
|
91 |
|
92 |
def save_uploaded_file(file):
|
93 |
if file is None:
|
|
|
128 |
|
129 |
return temp_path
|
130 |
|
131 |
+
das_pipeline = None
|
132 |
+
moge_model = None
|
133 |
+
|
134 |
+
def get_das_pipeline():
|
135 |
+
global das_pipeline
|
136 |
+
if das_pipeline is None:
|
137 |
+
das_pipeline = DiffusionAsShaderPipeline(gpu_id=GPU_ID, output_dir=OUTPUT_DIR)
|
138 |
+
return das_pipeline
|
139 |
+
|
140 |
+
def get_moge_model():
|
141 |
+
global moge_model
|
142 |
+
if moge_model is None:
|
143 |
+
das = get_das_pipeline()
|
144 |
+
moge_model = MoGeModel.from_pretrained("Ruicheng/moge-vitl").to(das.device)
|
145 |
+
return moge_model
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
146 |
|
|
|
147 |
def process_motion_transfer(source, prompt, mt_repaint_option, mt_repaint_image):
|
148 |
"""Process video motion transfer task"""
|
149 |
try:
|
|
|
155 |
print(f"DEBUG: Repaint option: {mt_repaint_option}")
|
156 |
print(f"DEBUG: Repaint image: {mt_repaint_image}")
|
157 |
|
158 |
+
|
159 |
+
das = get_das_pipeline()
|
160 |
+
video_tensor, fps, is_video = load_media(input_video_path)
|
161 |
+
if not is_video:
|
162 |
+
tracking_method = "moge"
|
163 |
+
print("Image input detected, using MoGe for tracking video generation.")
|
164 |
+
else:
|
165 |
+
tracking_method = "spatracker"
|
166 |
|
167 |
+
repaint_img_tensor = None
|
168 |
if mt_repaint_image is not None:
|
|
|
169 |
repaint_path = save_uploaded_file(mt_repaint_image)
|
170 |
+
repaint_img_tensor, _, _ = load_media(repaint_path)
|
171 |
+
repaint_img_tensor = repaint_img_tensor[0]
|
172 |
elif mt_repaint_option == "Yes":
|
173 |
+
repainter = FirstFrameRepainter(gpu_id=GPU_ID, output_dir=OUTPUT_DIR)
|
174 |
+
repaint_img_tensor = repainter.repaint(
|
175 |
+
video_tensor[0],
|
176 |
+
prompt=prompt,
|
177 |
+
depth_path=None
|
178 |
+
)
|
179 |
+
tracking_tensor = None
|
180 |
+
if tracking_method == "moge":
|
181 |
+
moge = get_moge_model()
|
182 |
+
infer_result = moge.infer(video_tensor[0].to(das.device)) # [C, H, W] in range [0,1]
|
183 |
+
H, W = infer_result["points"].shape[0:2]
|
184 |
+
pred_tracks = infer_result["points"].unsqueeze(0).repeat(49, 1, 1, 1) #[T, H, W, 3]
|
185 |
+
poses = torch.eye(4).unsqueeze(0).repeat(49, 1, 1)
|
186 |
+
|
187 |
+
pred_tracks_flatten = pred_tracks.reshape(video_tensor.shape[0], H*W, 3)
|
188 |
+
|
189 |
+
cam_motion = CameraMotionGenerator(None)
|
190 |
+
cam_motion.set_intr(infer_result["intrinsics"])
|
191 |
+
|
192 |
+
pred_tracks = cam_motion.w2s(pred_tracks_flatten, poses).reshape([video_tensor.shape[0], H, W, 3]) # [T, H, W, 3]
|
193 |
+
|
194 |
+
_, tracking_tensor = das.visualize_tracking_moge(
|
195 |
+
pred_tracks.cpu().numpy(),
|
196 |
+
infer_result["mask"].cpu().numpy()
|
197 |
+
)
|
198 |
+
print('Export tracking video via MoGe')
|
199 |
else:
|
200 |
+
pred_tracks, pred_visibility, T_Firsts = das.generate_tracking_spatracker(video_tensor)
|
201 |
+
|
202 |
+
_, tracking_tensor = das.visualize_tracking_spatracker(video_tensor, pred_tracks, pred_visibility, T_Firsts)
|
203 |
+
print('Export tracking video via SpaTracker')
|
204 |
+
|
205 |
+
output_path = das.apply_tracking(
|
206 |
+
video_tensor=video_tensor,
|
207 |
+
fps=8,
|
208 |
+
tracking_tensor=tracking_tensor,
|
209 |
+
img_cond_tensor=repaint_img_tensor,
|
210 |
+
prompt=prompt,
|
211 |
+
checkpoint_path=DEFAULT_MODEL_PATH
|
212 |
+
)
|
213 |
+
|
214 |
+
return output_path
|
215 |
except Exception as e:
|
216 |
import traceback
|
217 |
print(f"Processing failed: {str(e)}\n{traceback.format_exc()}")
|
218 |
return None
|
219 |
|
|
|
220 |
def process_camera_control(source, prompt, camera_motion, tracking_method):
|
221 |
"""Process camera control task"""
|
222 |
try:
|
|
|
228 |
print(f"DEBUG: Camera motion: '{camera_motion}'")
|
229 |
print(f"DEBUG: Tracking method: '{tracking_method}'")
|
230 |
|
231 |
+
das = get_das_pipeline()
|
232 |
+
|
233 |
+
video_tensor, fps, is_video = load_media(input_media_path)
|
234 |
+
if not is_video and tracking_method == "spatracker":
|
235 |
+
tracking_method = "moge"
|
236 |
+
print("Image input detected with spatracker selected, switching to MoGe")
|
237 |
+
|
238 |
+
cam_motion = CameraMotionGenerator(camera_motion)
|
239 |
+
repaint_img_tensor = None
|
240 |
+
tracking_tensor = None
|
241 |
+
|
242 |
+
if tracking_method == "moge":
|
243 |
+
moge = get_moge_model()
|
244 |
+
|
245 |
+
infer_result = moge.infer(video_tensor[0].to(das.device)) # [C, H, W] in range [0,1]
|
246 |
+
H, W = infer_result["points"].shape[0:2]
|
247 |
+
pred_tracks = infer_result["points"].unsqueeze(0).repeat(49, 1, 1, 1) #[T, H, W, 3]
|
248 |
+
cam_motion.set_intr(infer_result["intrinsics"])
|
249 |
+
|
250 |
+
if camera_motion:
|
251 |
+
poses = cam_motion.get_default_motion() # shape: [49, 4, 4]
|
252 |
+
print("Camera motion applied")
|
253 |
+
else:
|
254 |
+
poses = torch.eye(4).unsqueeze(0).repeat(49, 1, 1)
|
255 |
+
|
256 |
+
pred_tracks_flatten = pred_tracks.reshape(video_tensor.shape[0], H*W, 3)
|
257 |
+
pred_tracks = cam_motion.w2s(pred_tracks_flatten, poses).reshape([video_tensor.shape[0], H, W, 3]) # [T, H, W, 3]
|
258 |
+
|
259 |
+
_, tracking_tensor = das.visualize_tracking_moge(
|
260 |
+
pred_tracks.cpu().numpy(),
|
261 |
+
infer_result["mask"].cpu().numpy()
|
262 |
+
)
|
263 |
+
print('Export tracking video via MoGe')
|
264 |
+
else:
|
265 |
+
|
266 |
+
pred_tracks, pred_visibility, T_Firsts = das.generate_tracking_spatracker(video_tensor)
|
267 |
+
if camera_motion:
|
268 |
+
poses = cam_motion.get_default_motion() # shape: [49, 4, 4]
|
269 |
+
pred_tracks = cam_motion.apply_motion_on_pts(pred_tracks, poses)
|
270 |
+
print("Camera motion applied")
|
271 |
+
|
272 |
+
_, tracking_tensor = das.visualize_tracking_spatracker(video_tensor, pred_tracks, pred_visibility, T_Firsts)
|
273 |
+
print('Export tracking video via SpaTracker')
|
274 |
|
|
|
|
|
275 |
|
276 |
+
output_path = das.apply_tracking(
|
277 |
+
video_tensor=video_tensor,
|
278 |
+
fps=8,
|
279 |
+
tracking_tensor=tracking_tensor,
|
280 |
+
img_cond_tensor=repaint_img_tensor,
|
281 |
+
prompt=prompt,
|
282 |
+
checkpoint_path=DEFAULT_MODEL_PATH
|
283 |
+
)
|
284 |
|
285 |
+
return output_path
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
286 |
except Exception as e:
|
287 |
import traceback
|
288 |
print(f"Processing failed: {str(e)}\n{traceback.format_exc()}")
|
289 |
return None
|
290 |
|
|
|
291 |
def process_object_manipulation(source, prompt, object_motion, object_mask, tracking_method):
|
292 |
"""Process object manipulation task"""
|
293 |
try:
|
|
|
297 |
return None
|
298 |
|
299 |
object_mask_path = save_uploaded_file(object_mask)
|
300 |
+
if object_mask_path is None:
|
301 |
+
print("Object mask not provided")
|
302 |
+
return None
|
303 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
304 |
|
305 |
+
das = get_das_pipeline()
|
306 |
+
video_tensor, fps, is_video = load_media(input_image_path)
|
307 |
+
if not is_video and tracking_method == "spatracker":
|
308 |
+
tracking_method = "moge"
|
309 |
+
print("Image input detected with spatracker selected, switching to MoGe")
|
310 |
+
|
311 |
|
312 |
+
mask_image = Image.open(object_mask_path).convert('L')
|
313 |
+
mask_image = transforms.Resize((480, 720))(mask_image)
|
314 |
+
mask = torch.from_numpy(np.array(mask_image) > 127)
|
315 |
+
|
316 |
+
motion_generator = ObjectMotionGenerator(device=das.device)
|
317 |
+
repaint_img_tensor = None
|
318 |
+
tracking_tensor = None
|
319 |
+
if tracking_method == "moge":
|
320 |
+
moge = get_moge_model()
|
321 |
+
|
322 |
+
|
323 |
+
infer_result = moge.infer(video_tensor[0].to(das.device)) # [C, H, W] in range [0,1]
|
324 |
+
H, W = infer_result["points"].shape[0:2]
|
325 |
+
pred_tracks = infer_result["points"].unsqueeze(0).repeat(49, 1, 1, 1) #[T, H, W, 3]
|
326 |
+
|
327 |
+
pred_tracks = motion_generator.apply_motion(
|
328 |
+
pred_tracks=pred_tracks,
|
329 |
+
mask=mask,
|
330 |
+
motion_type=object_motion,
|
331 |
+
distance=50,
|
332 |
+
num_frames=49,
|
333 |
+
tracking_method="moge"
|
334 |
+
)
|
335 |
+
print(f"Object motion '{object_motion}' applied using provided mask")
|
336 |
+
poses = torch.eye(4).unsqueeze(0).repeat(49, 1, 1)
|
337 |
+
pred_tracks_flatten = pred_tracks.reshape(video_tensor.shape[0], H*W, 3)
|
338 |
+
|
339 |
+
|
340 |
+
cam_motion = CameraMotionGenerator(None)
|
341 |
+
cam_motion.set_intr(infer_result["intrinsics"])
|
342 |
+
pred_tracks = cam_motion.w2s(pred_tracks_flatten, poses).reshape([video_tensor.shape[0], H, W, 3]) # [T, H, W, 3]
|
343 |
+
|
344 |
+
_, tracking_tensor = das.visualize_tracking_moge(
|
345 |
+
pred_tracks.cpu().numpy(),
|
346 |
+
infer_result["mask"].cpu().numpy()
|
347 |
+
)
|
348 |
+
print('Export tracking video via MoGe')
|
349 |
else:
|
350 |
+
|
351 |
+
pred_tracks, pred_visibility, T_Firsts = das.generate_tracking_spatracker(video_tensor)
|
352 |
+
|
353 |
+
|
354 |
+
pred_tracks = motion_generator.apply_motion(
|
355 |
+
pred_tracks=pred_tracks.squeeze(),
|
356 |
+
mask=mask,
|
357 |
+
motion_type=object_motion,
|
358 |
+
distance=50,
|
359 |
+
num_frames=49,
|
360 |
+
tracking_method="spatracker"
|
361 |
+
).unsqueeze(0)
|
362 |
+
print(f"Object motion '{object_motion}' applied using provided mask")
|
363 |
+
|
364 |
+
|
365 |
+
_, tracking_tensor = das.visualize_tracking_spatracker(video_tensor, pred_tracks, pred_visibility, T_Firsts)
|
366 |
+
print('Export tracking video via SpaTracker')
|
367 |
+
|
368 |
+
|
369 |
+
output_path = das.apply_tracking(
|
370 |
+
video_tensor=video_tensor,
|
371 |
+
fps=8,
|
372 |
+
tracking_tensor=tracking_tensor,
|
373 |
+
img_cond_tensor=repaint_img_tensor,
|
374 |
+
prompt=prompt,
|
375 |
+
checkpoint_path=DEFAULT_MODEL_PATH
|
376 |
+
)
|
377 |
+
|
378 |
+
return output_path
|
379 |
except Exception as e:
|
380 |
import traceback
|
381 |
print(f"Processing failed: {str(e)}\n{traceback.format_exc()}")
|
382 |
return None
|
383 |
|
|
|
384 |
def process_mesh_animation(source, prompt, tracking_video, ma_repaint_option, ma_repaint_image):
|
385 |
"""Process mesh animation task"""
|
386 |
try:
|
|
|
393 |
if tracking_video_path is None:
|
394 |
return None
|
395 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
396 |
|
397 |
+
das = get_das_pipeline()
|
398 |
+
video_tensor, fps, is_video = load_media(input_video_path)
|
399 |
+
tracking_tensor, tracking_fps, _ = load_media(tracking_video_path)
|
400 |
+
repaint_img_tensor = None
|
401 |
if ma_repaint_image is not None:
|
|
|
402 |
repaint_path = save_uploaded_file(ma_repaint_image)
|
403 |
+
repaint_img_tensor, _, _ = load_media(repaint_path)
|
404 |
+
repaint_img_tensor = repaint_img_tensor[0] # 获取第一帧
|
405 |
elif ma_repaint_option == "Yes":
|
406 |
+
|
407 |
+
repainter = FirstFrameRepainter(gpu_id=GPU_ID, output_dir=OUTPUT_DIR)
|
408 |
+
repaint_img_tensor = repainter.repaint(
|
409 |
+
video_tensor[0],
|
410 |
+
prompt=prompt,
|
411 |
+
depth_path=None
|
412 |
+
)
|
413 |
+
|
414 |
+
output_path = das.apply_tracking(
|
415 |
+
video_tensor=video_tensor,
|
416 |
+
fps=8,
|
417 |
+
tracking_tensor=tracking_tensor,
|
418 |
+
img_cond_tensor=repaint_img_tensor,
|
419 |
+
prompt=prompt,
|
420 |
+
checkpoint_path=DEFAULT_MODEL_PATH
|
421 |
+
)
|
422 |
|
423 |
+
return output_path
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
424 |
except Exception as e:
|
425 |
import traceback
|
426 |
print(f"Processing failed: {str(e)}\n{traceback.format_exc()}")
|