Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -1,7 +1,6 @@
|
|
1 |
import gradio as gr
|
2 |
import spaces
|
3 |
from gradio_litmodel3d import LitModel3D
|
4 |
-
|
5 |
import os
|
6 |
import shutil
|
7 |
os.environ['SPCONV_ALGO'] = 'native'
|
@@ -15,32 +14,33 @@ from trellis.pipelines import TrellisImageTo3DPipeline
|
|
15 |
from trellis.representations import Gaussian, MeshExtractResult
|
16 |
from trellis.utils import render_utils, postprocessing_utils
|
17 |
|
18 |
-
|
19 |
MAX_SEED = np.iinfo(np.int32).max
|
20 |
TMP_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'tmp')
|
21 |
os.makedirs(TMP_DIR, exist_ok=True)
|
22 |
|
23 |
-
|
24 |
def start_session(req: gr.Request):
|
25 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
26 |
os.makedirs(user_dir, exist_ok=True)
|
27 |
-
|
28 |
-
|
29 |
def end_session(req: gr.Request):
|
30 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
31 |
shutil.rmtree(user_dir)
|
32 |
|
33 |
-
|
34 |
def preprocess_images(images: List[Tuple[Image.Image, str]]) -> List[Image.Image]:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
images = [image[0] for image in images]
|
36 |
processed_images = [pipeline.preprocess_image(image) for image in images]
|
37 |
return processed_images
|
38 |
|
39 |
-
|
40 |
def pack_state(gs: Gaussian, mesh: MeshExtractResult) -> dict:
|
41 |
return {
|
42 |
'gaussian': {
|
43 |
-
**gs.init_params,
|
44 |
'_xyz': gs._xyz.cpu().numpy(),
|
45 |
'_features_dc': gs._features_dc.cpu().numpy(),
|
46 |
'_scaling': gs._scaling.cpu().numpy(),
|
@@ -52,9 +52,8 @@ def pack_state(gs: Gaussian, mesh: MeshExtractResult) -> dict:
|
|
52 |
'faces': mesh.faces.cpu().numpy(),
|
53 |
},
|
54 |
}
|
55 |
-
|
56 |
-
|
57 |
-
def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
|
58 |
gs = Gaussian(
|
59 |
aabb=state['gaussian']['aabb'],
|
60 |
sh_degree=state['gaussian']['sh_degree'],
|
@@ -68,19 +67,18 @@ def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
|
|
68 |
gs._scaling = torch.tensor(state['gaussian']['_scaling'], device='cuda')
|
69 |
gs._rotation = torch.tensor(state['gaussian']['_rotation'], device='cuda')
|
70 |
gs._opacity = torch.tensor(state['gaussian']['_opacity'], device='cuda')
|
71 |
-
|
72 |
mesh = edict(
|
73 |
vertices=torch.tensor(state['mesh']['vertices'], device='cuda'),
|
74 |
faces=torch.tensor(state['mesh']['faces'], device='cuda'),
|
75 |
)
|
76 |
-
|
77 |
return gs, mesh
|
78 |
|
79 |
-
|
80 |
def get_seed(randomize_seed: bool, seed: int) -> int:
|
|
|
|
|
|
|
81 |
return np.random.randint(0, MAX_SEED) if randomize_seed else seed
|
82 |
|
83 |
-
|
84 |
@spaces.GPU
|
85 |
def image_to_3d(
|
86 |
multiimages: List[Tuple[Image.Image, str]],
|
@@ -92,22 +90,25 @@ def image_to_3d(
|
|
92 |
multiimage_algo: Literal["multidiffusion", "stochastic"],
|
93 |
req: gr.Request,
|
94 |
) -> Tuple[dict, str]:
|
|
|
|
|
|
|
95 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
video = render_utils.render_video(outputs['gaussian'][0], num_frames=120)['color']
|
112 |
video_geo = render_utils.render_video(outputs['mesh'][0], num_frames=120)['normal']
|
113 |
video = [np.concatenate([video[i], video_geo[i]], axis=1) for i in range(len(video))]
|
@@ -117,7 +118,6 @@ def image_to_3d(
|
|
117 |
torch.cuda.empty_cache()
|
118 |
return state, video_path
|
119 |
|
120 |
-
|
121 |
@spaces.GPU(duration=90)
|
122 |
def extract_glb(
|
123 |
state: dict,
|
@@ -125,6 +125,9 @@ def extract_glb(
|
|
125 |
texture_size: int,
|
126 |
req: gr.Request,
|
127 |
) -> Tuple[str, str]:
|
|
|
|
|
|
|
128 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
129 |
gs, mesh = unpack_state(state)
|
130 |
glb = postprocessing_utils.to_glb(gs, mesh, simplify=mesh_simplify, texture_size=texture_size, verbose=False)
|
@@ -133,9 +136,11 @@ def extract_glb(
|
|
133 |
torch.cuda.empty_cache()
|
134 |
return glb_path, glb_path
|
135 |
|
136 |
-
|
137 |
@spaces.GPU
|
138 |
def extract_gaussian(state: dict, req: gr.Request) -> Tuple[str, str]:
|
|
|
|
|
|
|
139 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
140 |
gs, _ = unpack_state(state)
|
141 |
gaussian_path = os.path.join(user_dir, 'sample.ply')
|
@@ -143,18 +148,17 @@ def extract_gaussian(state: dict, req: gr.Request) -> Tuple[str, str]:
|
|
143 |
torch.cuda.empty_cache()
|
144 |
return gaussian_path, gaussian_path
|
145 |
|
146 |
-
|
147 |
def prepare_multi_example() -> List[Image.Image]:
|
148 |
multi_case = list(set([i.split('_')[0] for i in os.listdir("assets/example_multi_image")]))
|
149 |
images = []
|
150 |
for case in multi_case:
|
151 |
-
|
152 |
for i in range(1, 4):
|
153 |
img = Image.open(f'assets/example_multi_image/{case}_{i}.png')
|
154 |
W, H = img.size
|
155 |
img = img.resize((int(W / H * 512), 512))
|
156 |
-
|
157 |
-
images.append(
|
158 |
return images
|
159 |
|
160 |
with gr.Blocks(delete_cache=(600, 600)) as demo:
|
@@ -208,30 +212,30 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
208 |
with gr.Row():
|
209 |
download_glb = gr.DownloadButton(label="Download GLB", interactive=False)
|
210 |
download_gs = gr.DownloadButton(label="Download Gaussian", interactive=False)
|
211 |
-
|
212 |
output_buf = gr.State()
|
213 |
-
|
214 |
# Example images at the bottom of the page
|
215 |
with gr.Row(visible=True) as multiimage_example:
|
216 |
examples_multi = gr.Examples(
|
217 |
examples=prepare_multi_example(),
|
218 |
inputs=[multiimage_prompt],
|
219 |
-
fn=
|
220 |
outputs=[multiimage_prompt],
|
221 |
run_on_click=True,
|
222 |
examples_per_page=8,
|
223 |
)
|
224 |
-
|
225 |
# Handlers
|
226 |
demo.load(start_session)
|
227 |
demo.unload(end_session)
|
228 |
-
|
229 |
multiimage_prompt.upload(
|
230 |
preprocess_images,
|
231 |
inputs=[multiimage_prompt],
|
232 |
outputs=[multiimage_prompt],
|
233 |
)
|
234 |
-
|
235 |
generate_btn.click(
|
236 |
get_seed,
|
237 |
inputs=[randomize_seed, seed],
|
@@ -244,12 +248,12 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
244 |
lambda: tuple([gr.Button(interactive=True), gr.Button(interactive=True)]),
|
245 |
outputs=[extract_glb_btn, extract_gs_btn],
|
246 |
)
|
247 |
-
|
248 |
video_output.clear(
|
249 |
lambda: tuple([gr.Button(interactive=False), gr.Button(interactive=False)]),
|
250 |
outputs=[extract_glb_btn, extract_gs_btn],
|
251 |
)
|
252 |
-
|
253 |
extract_glb_btn.click(
|
254 |
extract_glb,
|
255 |
inputs=[output_buf, mesh_simplify, texture_size],
|
@@ -258,7 +262,7 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
258 |
lambda: gr.Button(interactive=True),
|
259 |
outputs=[download_glb],
|
260 |
)
|
261 |
-
|
262 |
extract_gs_btn.click(
|
263 |
extract_gaussian,
|
264 |
inputs=[output_buf],
|
@@ -267,7 +271,7 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
267 |
lambda: gr.Button(interactive=True),
|
268 |
outputs=[download_gs],
|
269 |
)
|
270 |
-
|
271 |
model_output.clear(
|
272 |
lambda: gr.Button(interactive=False),
|
273 |
outputs=[download_glb],
|
@@ -281,4 +285,4 @@ if __name__ == "__main__":
|
|
281 |
pipeline.preprocess_image(Image.fromarray(np.zeros((512, 512, 3), dtype=np.uint8))) # Preload rembg
|
282 |
except:
|
283 |
pass
|
284 |
-
demo.launch(show_error=True)
|
|
|
1 |
import gradio as gr
|
2 |
import spaces
|
3 |
from gradio_litmodel3d import LitModel3D
|
|
|
4 |
import os
|
5 |
import shutil
|
6 |
os.environ['SPCONV_ALGO'] = 'native'
|
|
|
14 |
from trellis.representations import Gaussian, MeshExtractResult
|
15 |
from trellis.utils import render_utils, postprocessing_utils
|
16 |
|
|
|
17 |
MAX_SEED = np.iinfo(np.int32).max
|
18 |
TMP_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'tmp')
|
19 |
os.makedirs(TMP_DIR, exist_ok=True)
|
20 |
|
|
|
21 |
def start_session(req: gr.Request):
|
22 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
23 |
os.makedirs(user_dir, exist_ok=True)
|
24 |
+
|
|
|
25 |
def end_session(req: gr.Request):
|
26 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
27 |
shutil.rmtree(user_dir)
|
28 |
|
|
|
29 |
def preprocess_images(images: List[Tuple[Image.Image, str]]) -> List[Image.Image]:
|
30 |
+
"""
|
31 |
+
Preprocess a list of input images.
|
32 |
+
Args:
|
33 |
+
images (List[Tuple[Image.Image, str]]): The input images.
|
34 |
+
Returns:
|
35 |
+
List[Image.Image]: The preprocessed images.
|
36 |
+
"""
|
37 |
images = [image[0] for image in images]
|
38 |
processed_images = [pipeline.preprocess_image(image) for image in images]
|
39 |
return processed_images
|
40 |
|
|
|
41 |
def pack_state(gs: Gaussian, mesh: MeshExtractResult) -> dict:
|
42 |
return {
|
43 |
'gaussian': {
|
|
|
44 |
'_xyz': gs._xyz.cpu().numpy(),
|
45 |
'_features_dc': gs._features_dc.cpu().numpy(),
|
46 |
'_scaling': gs._scaling.cpu().numpy(),
|
|
|
52 |
'faces': mesh.faces.cpu().numpy(),
|
53 |
},
|
54 |
}
|
55 |
+
|
56 |
+
def unpack_state(state: dict) -> Tuple[Gaussian, edict]:
|
|
|
57 |
gs = Gaussian(
|
58 |
aabb=state['gaussian']['aabb'],
|
59 |
sh_degree=state['gaussian']['sh_degree'],
|
|
|
67 |
gs._scaling = torch.tensor(state['gaussian']['_scaling'], device='cuda')
|
68 |
gs._rotation = torch.tensor(state['gaussian']['_rotation'], device='cuda')
|
69 |
gs._opacity = torch.tensor(state['gaussian']['_opacity'], device='cuda')
|
|
|
70 |
mesh = edict(
|
71 |
vertices=torch.tensor(state['mesh']['vertices'], device='cuda'),
|
72 |
faces=torch.tensor(state['mesh']['faces'], device='cuda'),
|
73 |
)
|
|
|
74 |
return gs, mesh
|
75 |
|
|
|
76 |
def get_seed(randomize_seed: bool, seed: int) -> int:
|
77 |
+
"""
|
78 |
+
Get the random seed.
|
79 |
+
"""
|
80 |
return np.random.randint(0, MAX_SEED) if randomize_seed else seed
|
81 |
|
|
|
82 |
@spaces.GPU
|
83 |
def image_to_3d(
|
84 |
multiimages: List[Tuple[Image.Image, str]],
|
|
|
90 |
multiimage_algo: Literal["multidiffusion", "stochastic"],
|
91 |
req: gr.Request,
|
92 |
) -> Tuple[dict, str]:
|
93 |
+
"""
|
94 |
+
Convert multiple images to a 3D model.
|
95 |
+
"""
|
96 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
97 |
+
outputs = pipeline.run_multi_image(
|
98 |
+
[image[0] for image in multiimages],
|
99 |
+
seed=seed,
|
100 |
+
formats=["gaussian", "mesh"],
|
101 |
+
preprocess_image=False,
|
102 |
+
sparse_structure_sampler_params={
|
103 |
+
"steps": ss_sampling_steps,
|
104 |
+
"cfg_strength": ss_guidance_strength,
|
105 |
+
},
|
106 |
+
slat_sampler_params={
|
107 |
+
"steps": slat_sampling_steps,
|
108 |
+
"cfg_strength": slat_guidance_strength,
|
109 |
+
},
|
110 |
+
mode=multiimage_algo,
|
111 |
+
)
|
112 |
video = render_utils.render_video(outputs['gaussian'][0], num_frames=120)['color']
|
113 |
video_geo = render_utils.render_video(outputs['mesh'][0], num_frames=120)['normal']
|
114 |
video = [np.concatenate([video[i], video_geo[i]], axis=1) for i in range(len(video))]
|
|
|
118 |
torch.cuda.empty_cache()
|
119 |
return state, video_path
|
120 |
|
|
|
121 |
@spaces.GPU(duration=90)
|
122 |
def extract_glb(
|
123 |
state: dict,
|
|
|
125 |
texture_size: int,
|
126 |
req: gr.Request,
|
127 |
) -> Tuple[str, str]:
|
128 |
+
"""
|
129 |
+
Extract a GLB file from the 3D model.
|
130 |
+
"""
|
131 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
132 |
gs, mesh = unpack_state(state)
|
133 |
glb = postprocessing_utils.to_glb(gs, mesh, simplify=mesh_simplify, texture_size=texture_size, verbose=False)
|
|
|
136 |
torch.cuda.empty_cache()
|
137 |
return glb_path, glb_path
|
138 |
|
|
|
139 |
@spaces.GPU
|
140 |
def extract_gaussian(state: dict, req: gr.Request) -> Tuple[str, str]:
|
141 |
+
"""
|
142 |
+
Extract a Gaussian file from the 3D model.
|
143 |
+
"""
|
144 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
145 |
gs, _ = unpack_state(state)
|
146 |
gaussian_path = os.path.join(user_dir, 'sample.ply')
|
|
|
148 |
torch.cuda.empty_cache()
|
149 |
return gaussian_path, gaussian_path
|
150 |
|
|
|
151 |
def prepare_multi_example() -> List[Image.Image]:
|
152 |
multi_case = list(set([i.split('_')[0] for i in os.listdir("assets/example_multi_image")]))
|
153 |
images = []
|
154 |
for case in multi_case:
|
155 |
+
views = []
|
156 |
for i in range(1, 4):
|
157 |
img = Image.open(f'assets/example_multi_image/{case}_{i}.png')
|
158 |
W, H = img.size
|
159 |
img = img.resize((int(W / H * 512), 512))
|
160 |
+
views.append(img)
|
161 |
+
images.append(views)
|
162 |
return images
|
163 |
|
164 |
with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
|
212 |
with gr.Row():
|
213 |
download_glb = gr.DownloadButton(label="Download GLB", interactive=False)
|
214 |
download_gs = gr.DownloadButton(label="Download Gaussian", interactive=False)
|
215 |
+
|
216 |
output_buf = gr.State()
|
217 |
+
|
218 |
# Example images at the bottom of the page
|
219 |
with gr.Row(visible=True) as multiimage_example:
|
220 |
examples_multi = gr.Examples(
|
221 |
examples=prepare_multi_example(),
|
222 |
inputs=[multiimage_prompt],
|
223 |
+
fn=lambda x: x,
|
224 |
outputs=[multiimage_prompt],
|
225 |
run_on_click=True,
|
226 |
examples_per_page=8,
|
227 |
)
|
228 |
+
|
229 |
# Handlers
|
230 |
demo.load(start_session)
|
231 |
demo.unload(end_session)
|
232 |
+
|
233 |
multiimage_prompt.upload(
|
234 |
preprocess_images,
|
235 |
inputs=[multiimage_prompt],
|
236 |
outputs=[multiimage_prompt],
|
237 |
)
|
238 |
+
|
239 |
generate_btn.click(
|
240 |
get_seed,
|
241 |
inputs=[randomize_seed, seed],
|
|
|
248 |
lambda: tuple([gr.Button(interactive=True), gr.Button(interactive=True)]),
|
249 |
outputs=[extract_glb_btn, extract_gs_btn],
|
250 |
)
|
251 |
+
|
252 |
video_output.clear(
|
253 |
lambda: tuple([gr.Button(interactive=False), gr.Button(interactive=False)]),
|
254 |
outputs=[extract_glb_btn, extract_gs_btn],
|
255 |
)
|
256 |
+
|
257 |
extract_glb_btn.click(
|
258 |
extract_glb,
|
259 |
inputs=[output_buf, mesh_simplify, texture_size],
|
|
|
262 |
lambda: gr.Button(interactive=True),
|
263 |
outputs=[download_glb],
|
264 |
)
|
265 |
+
|
266 |
extract_gs_btn.click(
|
267 |
extract_gaussian,
|
268 |
inputs=[output_buf],
|
|
|
271 |
lambda: gr.Button(interactive=True),
|
272 |
outputs=[download_gs],
|
273 |
)
|
274 |
+
|
275 |
model_output.clear(
|
276 |
lambda: gr.Button(interactive=False),
|
277 |
outputs=[download_glb],
|
|
|
285 |
pipeline.preprocess_image(Image.fromarray(np.zeros((512, 512, 3), dtype=np.uint8))) # Preload rembg
|
286 |
except:
|
287 |
pass
|
288 |
+
demo.launch(show_error=True)
|