Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -13,7 +13,6 @@ from PIL import Image
|
|
13 |
from trellis.pipelines import TrellisImageTo3DPipeline
|
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)
|
@@ -28,9 +27,6 @@ def end_session(req: gr.Request):
|
|
28 |
shutil.rmtree(user_dir)
|
29 |
|
30 |
def preprocess_images(images: List[Tuple[Image.Image, str]]) -> List[Image.Image]:
|
31 |
-
"""
|
32 |
-
Preprocesa una lista de im谩genes.
|
33 |
-
"""
|
34 |
images = [image[0] for image in images]
|
35 |
processed_images = [pipeline.preprocess_image(image) for image in images]
|
36 |
return processed_images
|
@@ -72,9 +68,6 @@ def unpack_state(state: dict) -> Tuple[Gaussian, edict]:
|
|
72 |
return gs, mesh
|
73 |
|
74 |
def get_seed(randomize_seed: bool, seed: int) -> int:
|
75 |
-
"""
|
76 |
-
Obtiene una semilla aleatoria.
|
77 |
-
"""
|
78 |
return np.random.randint(0, MAX_SEED) if randomize_seed else seed
|
79 |
|
80 |
@spaces.GPU
|
@@ -88,9 +81,6 @@ def image_to_3d(
|
|
88 |
multiimage_algo: Literal["multidiffusion", "stochastic"],
|
89 |
req: gr.Request,
|
90 |
) -> Tuple[dict, str]:
|
91 |
-
"""
|
92 |
-
Convierte m煤ltiples im谩genes en un modelo 3D.
|
93 |
-
"""
|
94 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
95 |
outputs = pipeline.run_multi_image(
|
96 |
[image[0] for image in multiimages],
|
@@ -123,9 +113,6 @@ def extract_glb(
|
|
123 |
texture_size: int,
|
124 |
req: gr.Request,
|
125 |
) -> Tuple[str, str]:
|
126 |
-
"""
|
127 |
-
Extrae un archivo GLB del modelo 3D.
|
128 |
-
"""
|
129 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
130 |
gs, mesh = unpack_state(state)
|
131 |
glb = postprocessing_utils.to_glb(gs, mesh, simplify=mesh_simplify, texture_size=texture_size, verbose=False)
|
@@ -134,41 +121,14 @@ def extract_glb(
|
|
134 |
torch.cuda.empty_cache()
|
135 |
return glb_path, glb_path
|
136 |
|
137 |
-
@spaces.GPU
|
138 |
-
def extract_gaussian(state: dict, req: gr.Request) -> Tuple[str, str]:
|
139 |
-
"""
|
140 |
-
Extrae un archivo Gaussiano del modelo 3D.
|
141 |
-
"""
|
142 |
-
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
143 |
-
gs, _ = unpack_state(state)
|
144 |
-
gaussian_path = os.path.join(user_dir, 'sample.ply')
|
145 |
-
gs.save_ply(gaussian_path)
|
146 |
-
torch.cuda.empty_cache()
|
147 |
-
return gaussian_path, gaussian_path
|
148 |
-
|
149 |
-
def prepare_multi_example() -> List[Tuple[str, str]]:
|
150 |
-
"""
|
151 |
-
Prepara ejemplos de m煤ltiples im谩genes para la galer铆a.
|
152 |
-
"""
|
153 |
-
multi_case = list(set([i.split('_')[0] for i in os.listdir("assets/example_multi_image")]))
|
154 |
-
examples = []
|
155 |
-
for case in multi_case:
|
156 |
-
case_images = []
|
157 |
-
for i in range(1, 4): # Suponemos 3 vistas por caso
|
158 |
-
img_path = f'assets/example_multi_image/{case}_{i}.png'
|
159 |
-
if os.path.exists(img_path): # Asegurarse de que la imagen existe
|
160 |
-
case_images.append((img_path, f"View {i}"))
|
161 |
-
if case_images: # Solo a帽adir casos con im谩genes v谩lidas
|
162 |
-
examples.append(case_images)
|
163 |
-
return examples
|
164 |
-
|
165 |
# Interfaz Gradio
|
166 |
with gr.Blocks(delete_cache=(600, 600)) as demo:
|
167 |
gr.Markdown("""
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
|
|
172 |
""")
|
173 |
|
174 |
with gr.Row():
|
@@ -176,11 +136,7 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
176 |
with gr.Tabs() as input_tabs:
|
177 |
with gr.Tab(label="Multiple Images", id=1) as multiimage_input_tab:
|
178 |
multiimage_prompt = gr.Gallery(label="Image Prompt", format="png", type="pil", height=300, columns=3)
|
179 |
-
|
180 |
-
Input different views of the object in separate images.
|
181 |
-
NOTE: this is an experimental algorithm without training a specialized model. It may not produce the best results for all images, especially those having different poses or inconsistent details.*
|
182 |
-
""")
|
183 |
-
|
184 |
with gr.Accordion(label="Generation Settings", open=False):
|
185 |
seed = gr.Slider(0, MAX_SEED, label="Seed", value=0, step=1)
|
186 |
randomize_seed = gr.Checkbox(label="Randomize Seed", value=True)
|
@@ -200,45 +156,25 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
200 |
mesh_simplify = gr.Slider(0.9, 0.98, label="Simplify", value=0.95, step=0.01)
|
201 |
texture_size = gr.Slider(512, 2048, label="Texture Size", value=1024, step=512)
|
202 |
|
203 |
-
|
204 |
-
extract_glb_btn = gr.Button("Extract GLB", interactive=False)
|
205 |
-
extract_gs_btn = gr.Button("Extract Gaussian", interactive=False)
|
206 |
-
|
207 |
-
gr.Markdown("""
|
208 |
-
NOTE: Gaussian file can be very large (~50MB), it will take a while to display and download.*
|
209 |
-
""")
|
210 |
|
211 |
with gr.Column():
|
212 |
video_output = gr.Video(label="Generated 3D Asset", autoplay=True, loop=True, height=300)
|
213 |
-
model_output = LitModel3D(label="Extracted GLB
|
214 |
-
|
215 |
-
with gr.Row():
|
216 |
-
download_glb = gr.DownloadButton(label="Download GLB", interactive=False)
|
217 |
-
download_gs = gr.DownloadButton(label="Download Gaussian", interactive=False)
|
218 |
|
219 |
output_buf = gr.State()
|
220 |
|
221 |
-
# Ejemplos de im谩genes m煤ltiples
|
222 |
-
with gr.Row(visible=True) as multiimage_example:
|
223 |
-
examples_multi = gr.Examples(
|
224 |
-
examples=prepare_multi_example(),
|
225 |
-
inputs=[multiimage_prompt],
|
226 |
-
fn=lambda x: x, # Pasar la entrada directamente (sin preprocesamiento adicional)
|
227 |
-
outputs=[multiimage_prompt],
|
228 |
-
run_on_click=True,
|
229 |
-
examples_per_page=8,
|
230 |
-
)
|
231 |
-
|
232 |
# Manejadores
|
233 |
demo.load(start_session)
|
234 |
demo.unload(end_session)
|
235 |
-
|
236 |
multiimage_prompt.upload(
|
237 |
preprocess_images,
|
238 |
inputs=[multiimage_prompt],
|
239 |
outputs=[multiimage_prompt],
|
240 |
)
|
241 |
-
|
242 |
generate_btn.click(
|
243 |
get_seed,
|
244 |
inputs=[randomize_seed, seed],
|
@@ -248,15 +184,15 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
248 |
inputs=[multiimage_prompt, seed, ss_guidance_strength, ss_sampling_steps, slat_guidance_strength, slat_sampling_steps, multiimage_algo],
|
249 |
outputs=[output_buf, video_output],
|
250 |
).then(
|
251 |
-
lambda:
|
252 |
-
outputs=[extract_glb_btn
|
253 |
)
|
254 |
-
|
255 |
video_output.clear(
|
256 |
-
lambda:
|
257 |
-
outputs=[extract_glb_btn
|
258 |
)
|
259 |
-
|
260 |
extract_glb_btn.click(
|
261 |
extract_glb,
|
262 |
inputs=[output_buf, mesh_simplify, texture_size],
|
@@ -265,16 +201,7 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
265 |
lambda: gr.Button(interactive=True),
|
266 |
outputs=[download_glb],
|
267 |
)
|
268 |
-
|
269 |
-
extract_gs_btn.click(
|
270 |
-
extract_gaussian,
|
271 |
-
inputs=[output_buf],
|
272 |
-
outputs=[model_output, download_gs],
|
273 |
-
).then(
|
274 |
-
lambda: gr.Button(interactive=True),
|
275 |
-
outputs=[download_gs],
|
276 |
-
)
|
277 |
-
|
278 |
model_output.clear(
|
279 |
lambda: gr.Button(interactive=False),
|
280 |
outputs=[download_glb],
|
|
|
13 |
from trellis.pipelines import TrellisImageTo3DPipeline
|
14 |
from trellis.representations import Gaussian, MeshExtractResult
|
15 |
from trellis.utils import render_utils, postprocessing_utils
|
|
|
16 |
MAX_SEED = np.iinfo(np.int32).max
|
17 |
TMP_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'tmp')
|
18 |
os.makedirs(TMP_DIR, exist_ok=True)
|
|
|
27 |
shutil.rmtree(user_dir)
|
28 |
|
29 |
def preprocess_images(images: List[Tuple[Image.Image, str]]) -> List[Image.Image]:
|
|
|
|
|
|
|
30 |
images = [image[0] for image in images]
|
31 |
processed_images = [pipeline.preprocess_image(image) for image in images]
|
32 |
return processed_images
|
|
|
68 |
return gs, mesh
|
69 |
|
70 |
def get_seed(randomize_seed: bool, seed: int) -> int:
|
|
|
|
|
|
|
71 |
return np.random.randint(0, MAX_SEED) if randomize_seed else seed
|
72 |
|
73 |
@spaces.GPU
|
|
|
81 |
multiimage_algo: Literal["multidiffusion", "stochastic"],
|
82 |
req: gr.Request,
|
83 |
) -> Tuple[dict, str]:
|
|
|
|
|
|
|
84 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
85 |
outputs = pipeline.run_multi_image(
|
86 |
[image[0] for image in multiimages],
|
|
|
113 |
texture_size: int,
|
114 |
req: gr.Request,
|
115 |
) -> Tuple[str, str]:
|
|
|
|
|
|
|
116 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
117 |
gs, mesh = unpack_state(state)
|
118 |
glb = postprocessing_utils.to_glb(gs, mesh, simplify=mesh_simplify, texture_size=texture_size, verbose=False)
|
|
|
121 |
torch.cuda.empty_cache()
|
122 |
return glb_path, glb_path
|
123 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
124 |
# Interfaz Gradio
|
125 |
with gr.Blocks(delete_cache=(600, 600)) as demo:
|
126 |
gr.Markdown("""
|
127 |
+
# UTPL - Conversi贸n de Multiples Im谩genes a objetos 3D usando IA
|
128 |
+
### Tesis: *"Objetos tridimensionales creados por IA: Innovaci贸n en entornos virtuales"*
|
129 |
+
**Autor:** Carlos Vargas
|
130 |
+
**Base t茅cnica:** Adaptaci贸n de [TRELLIS](https://trellis3d.github.io/) (herramienta de c贸digo abierto para generaci贸n 3D)
|
131 |
+
**Prop贸sito educativo:** Demostraciones acad茅micas e Investigaci贸n en modelado 3D autom谩tico
|
132 |
""")
|
133 |
|
134 |
with gr.Row():
|
|
|
136 |
with gr.Tabs() as input_tabs:
|
137 |
with gr.Tab(label="Multiple Images", id=1) as multiimage_input_tab:
|
138 |
multiimage_prompt = gr.Gallery(label="Image Prompt", format="png", type="pil", height=300, columns=3)
|
139 |
+
|
|
|
|
|
|
|
|
|
140 |
with gr.Accordion(label="Generation Settings", open=False):
|
141 |
seed = gr.Slider(0, MAX_SEED, label="Seed", value=0, step=1)
|
142 |
randomize_seed = gr.Checkbox(label="Randomize Seed", value=True)
|
|
|
156 |
mesh_simplify = gr.Slider(0.9, 0.98, label="Simplify", value=0.95, step=0.01)
|
157 |
texture_size = gr.Slider(512, 2048, label="Texture Size", value=1024, step=512)
|
158 |
|
159 |
+
extract_glb_btn = gr.Button("Extract GLB", interactive=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
160 |
|
161 |
with gr.Column():
|
162 |
video_output = gr.Video(label="Generated 3D Asset", autoplay=True, loop=True, height=300)
|
163 |
+
model_output = LitModel3D(label="Extracted GLB", exposure=10.0, height=300)
|
164 |
+
download_glb = gr.DownloadButton(label="Download GLB", interactive=False)
|
|
|
|
|
|
|
165 |
|
166 |
output_buf = gr.State()
|
167 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
168 |
# Manejadores
|
169 |
demo.load(start_session)
|
170 |
demo.unload(end_session)
|
171 |
+
|
172 |
multiimage_prompt.upload(
|
173 |
preprocess_images,
|
174 |
inputs=[multiimage_prompt],
|
175 |
outputs=[multiimage_prompt],
|
176 |
)
|
177 |
+
|
178 |
generate_btn.click(
|
179 |
get_seed,
|
180 |
inputs=[randomize_seed, seed],
|
|
|
184 |
inputs=[multiimage_prompt, seed, ss_guidance_strength, ss_sampling_steps, slat_guidance_strength, slat_sampling_steps, multiimage_algo],
|
185 |
outputs=[output_buf, video_output],
|
186 |
).then(
|
187 |
+
lambda: gr.Button(interactive=True),
|
188 |
+
outputs=[extract_glb_btn],
|
189 |
)
|
190 |
+
|
191 |
video_output.clear(
|
192 |
+
lambda: gr.Button(interactive=False),
|
193 |
+
outputs=[extract_glb_btn],
|
194 |
)
|
195 |
+
|
196 |
extract_glb_btn.click(
|
197 |
extract_glb,
|
198 |
inputs=[output_buf, mesh_simplify, texture_size],
|
|
|
201 |
lambda: gr.Button(interactive=True),
|
202 |
outputs=[download_glb],
|
203 |
)
|
204 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
205 |
model_output.clear(
|
206 |
lambda: gr.Button(interactive=False),
|
207 |
outputs=[download_glb],
|