Spaces:
Running
on
L40S
Running
on
L40S
Update app.py
Browse files
app.py
CHANGED
@@ -2,8 +2,6 @@ import gradio as gr
|
|
2 |
import spaces
|
3 |
from gradio_litmodel3d import LitModel3D
|
4 |
import os
|
5 |
-
os.environ['SPCONV_ALGO'] = 'native'
|
6 |
-
from typing import *
|
7 |
import torch
|
8 |
import numpy as np
|
9 |
import imageio
|
@@ -15,6 +13,11 @@ from trellis.representations import Gaussian, MeshExtractResult
|
|
15 |
from trellis.utils import render_utils, postprocessing_utils
|
16 |
from transformers import pipeline as translation_pipeline
|
17 |
from diffusers import FluxPipeline
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
MAX_SEED = np.iinfo(np.int32).max
|
20 |
TMP_DIR = "/tmp/Trellis-demo"
|
@@ -24,37 +27,27 @@ def initialize_models():
|
|
24 |
global pipeline, translator, flux_pipe
|
25 |
|
26 |
try:
|
27 |
-
#
|
28 |
-
torch.cuda.empty_cache()
|
29 |
-
|
30 |
-
# GPU ์ฌ์ฉ ๊ฐ๋ฅ ์ฌ๋ถ ํ์ธ
|
31 |
-
device = "cuda" if torch.cuda.is_available() else "cpu"
|
32 |
-
|
33 |
-
# Trellis ํ์ดํ๋ผ์ธ ์ด๊ธฐํ
|
34 |
pipeline = TrellisImageTo3DPipeline.from_pretrained("JeffreyXiang/TRELLIS-image-large")
|
35 |
-
pipeline.to(device)
|
36 |
|
37 |
-
# ๋ฒ์ญ๊ธฐ ์ด๊ธฐํ
|
38 |
translator = translation_pipeline(
|
39 |
"translation",
|
40 |
model="Helsinki-NLP/opus-mt-ko-en",
|
41 |
-
device
|
42 |
)
|
43 |
|
44 |
-
# Flux ํ์ดํ๋ผ์ธ ์ด๊ธฐํ
|
45 |
flux_pipe = FluxPipeline.from_pretrained(
|
46 |
"black-forest-labs/FLUX.1-dev",
|
47 |
-
torch_dtype=torch.
|
48 |
)
|
49 |
|
50 |
-
|
51 |
-
flux_pipe.enable_model_cpu_offload()
|
52 |
-
|
53 |
return True
|
54 |
|
55 |
except Exception as e:
|
56 |
print(f"Model initialization error: {str(e)}")
|
57 |
-
torch.cuda.empty_cache()
|
58 |
return False
|
59 |
|
60 |
def translate_if_korean(text):
|
@@ -63,11 +56,25 @@ def translate_if_korean(text):
|
|
63 |
return translated
|
64 |
return text
|
65 |
|
|
|
66 |
def preprocess_image(image: Image.Image) -> Tuple[str, Image.Image]:
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
|
72 |
def pack_state(gs: Gaussian, mesh: MeshExtractResult, trial_id: str) -> dict:
|
73 |
return {
|
@@ -86,7 +93,6 @@ def pack_state(gs: Gaussian, mesh: MeshExtractResult, trial_id: str) -> dict:
|
|
86 |
'trial_id': trial_id,
|
87 |
}
|
88 |
|
89 |
-
|
90 |
def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
|
91 |
gs = Gaussian(
|
92 |
aabb=state['gaussian']['aabb'],
|
@@ -113,31 +119,32 @@ def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
|
|
113 |
def image_to_3d(trial_id: str, seed: int, randomize_seed: bool, ss_guidance_strength: float,
|
114 |
ss_sampling_steps: int, slat_guidance_strength: float, slat_sampling_steps: int):
|
115 |
try:
|
116 |
-
torch.cuda.empty_cache()
|
117 |
-
|
118 |
if randomize_seed:
|
119 |
seed = np.random.randint(0, MAX_SEED)
|
120 |
|
121 |
input_image = Image.open(f"{TMP_DIR}/{trial_id}.png")
|
122 |
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
|
|
|
|
|
|
|
|
139 |
|
140 |
-
# ๋น๋์ค ๋ ๋๋ง
|
141 |
video = render_utils.render_video(outputs['gaussian'][0], num_frames=120)['color']
|
142 |
video_geo = render_utils.render_video(outputs['mesh'][0], num_frames=120)['normal']
|
143 |
video = [np.concatenate([video[i], video_geo[i]], axis=1) for i in range(len(video))]
|
@@ -149,37 +156,51 @@ def image_to_3d(trial_id: str, seed: int, randomize_seed: bool, ss_guidance_stre
|
|
149 |
|
150 |
state = pack_state(outputs['gaussian'][0], outputs['mesh'][0], trial_id)
|
151 |
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
return state, video_path
|
156 |
|
157 |
except Exception as e:
|
158 |
print(f"Error in image_to_3d: {str(e)}")
|
159 |
-
|
160 |
raise e
|
161 |
|
162 |
@spaces.GPU
|
163 |
def generate_image_from_text(prompt, height, width, guidance_scale, num_steps):
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
181 |
|
182 |
return image
|
|
|
|
|
|
|
|
|
|
|
183 |
|
184 |
@spaces.GPU
|
185 |
def extract_glb(state: dict, mesh_simplify: float, texture_size: int) -> Tuple[str, str]:
|
@@ -195,14 +216,13 @@ def activate_button() -> gr.Button:
|
|
195 |
def deactivate_button() -> gr.Button:
|
196 |
return gr.Button(interactive=False)
|
197 |
|
198 |
-
|
199 |
css = """
|
200 |
footer {
|
201 |
visibility: hidden;
|
202 |
}
|
203 |
"""
|
204 |
|
205 |
-
|
206 |
with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
|
207 |
gr.Markdown("""
|
208 |
# Craft3D : 3D Asset Creation & Text-to-Image Generation
|
@@ -278,7 +298,7 @@ with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
|
|
278 |
examples_per_page=64,
|
279 |
)
|
280 |
|
281 |
-
# Handlers
|
282 |
image_prompt.upload(
|
283 |
preprocess_image,
|
284 |
inputs=[image_prompt],
|
@@ -292,59 +312,48 @@ with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
|
|
292 |
|
293 |
generate_btn.click(
|
294 |
image_to_3d,
|
295 |
-
inputs=[trial_id, seed, randomize_seed, ss_guidance_strength, ss_sampling_steps,
|
|
|
296 |
outputs=[output_buf, video_output],
|
|
|
297 |
).then(
|
298 |
activate_button,
|
299 |
-
outputs=[extract_glb_btn]
|
300 |
-
)
|
301 |
-
|
302 |
-
video_output.clear(
|
303 |
-
deactivate_button,
|
304 |
-
outputs=[extract_glb_btn],
|
305 |
)
|
306 |
|
307 |
extract_glb_btn.click(
|
308 |
extract_glb,
|
309 |
inputs=[output_buf, mesh_simplify, texture_size],
|
310 |
outputs=[model_output, download_glb],
|
|
|
311 |
).then(
|
312 |
activate_button,
|
313 |
-
outputs=[download_glb]
|
314 |
)
|
315 |
|
316 |
-
model_output.clear(
|
317 |
-
deactivate_button,
|
318 |
-
outputs=[download_glb],
|
319 |
-
)
|
320 |
-
|
321 |
-
# Text to Image ํธ๋ค๋ฌ
|
322 |
generate_txt2img_btn.click(
|
323 |
generate_image_from_text,
|
324 |
inputs=[text_prompt, txt2img_height, txt2img_width, guidance_scale, num_steps],
|
325 |
-
outputs=[txt2img_output]
|
|
|
326 |
)
|
327 |
|
328 |
if __name__ == "__main__":
|
329 |
-
#
|
330 |
-
if torch.cuda.is_available():
|
331 |
-
torch.cuda.empty_cache()
|
332 |
-
|
333 |
-
# ๋ชจ๋ธ ์ด๊ธฐํ ํ์ธ
|
334 |
if not initialize_models():
|
335 |
print("Failed to initialize models")
|
336 |
exit(1)
|
337 |
|
338 |
try:
|
339 |
# rembg ์ฌ์ ๋ก๋ ์๋
|
340 |
-
test_image = Image.fromarray(np.
|
341 |
pipeline.preprocess_image(test_image)
|
342 |
except Exception as e:
|
343 |
print(f"Warning: Failed to preload rembg: {str(e)}")
|
344 |
|
345 |
# Gradio ์ฑ ์คํ
|
346 |
-
demo.queue(
|
347 |
share=True,
|
348 |
-
|
349 |
-
|
350 |
)
|
|
|
2 |
import spaces
|
3 |
from gradio_litmodel3d import LitModel3D
|
4 |
import os
|
|
|
|
|
5 |
import torch
|
6 |
import numpy as np
|
7 |
import imageio
|
|
|
13 |
from trellis.utils import render_utils, postprocessing_utils
|
14 |
from transformers import pipeline as translation_pipeline
|
15 |
from diffusers import FluxPipeline
|
16 |
+
from typing import *
|
17 |
+
|
18 |
+
# ํ๊ฒฝ ๋ณ์ ์ค์
|
19 |
+
os.environ['SPCONV_ALGO'] = 'native'
|
20 |
+
os.environ['WARP_USE_CPU'] = '1' # Warp๋ฅผ CPU ๋ชจ๋๋ก ๊ฐ์
|
21 |
|
22 |
MAX_SEED = np.iinfo(np.int32).max
|
23 |
TMP_DIR = "/tmp/Trellis-demo"
|
|
|
27 |
global pipeline, translator, flux_pipe
|
28 |
|
29 |
try:
|
30 |
+
# Trellis ํ์ดํ๋ผ์ธ ์ด๊ธฐํ (CPU ๋ชจ๋๋ก)
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
pipeline = TrellisImageTo3DPipeline.from_pretrained("JeffreyXiang/TRELLIS-image-large")
|
|
|
32 |
|
33 |
+
# ๋ฒ์ญ๊ธฐ ์ด๊ธฐํ (CPU ๋ชจ๋๋ก)
|
34 |
translator = translation_pipeline(
|
35 |
"translation",
|
36 |
model="Helsinki-NLP/opus-mt-ko-en",
|
37 |
+
device=-1
|
38 |
)
|
39 |
|
40 |
+
# Flux ํ์ดํ๋ผ์ธ ์ด๊ธฐํ (CPU ๋ชจ๋๋ก)
|
41 |
flux_pipe = FluxPipeline.from_pretrained(
|
42 |
"black-forest-labs/FLUX.1-dev",
|
43 |
+
torch_dtype=torch.float32
|
44 |
)
|
45 |
|
46 |
+
print("Models initialized successfully")
|
|
|
|
|
47 |
return True
|
48 |
|
49 |
except Exception as e:
|
50 |
print(f"Model initialization error: {str(e)}")
|
|
|
51 |
return False
|
52 |
|
53 |
def translate_if_korean(text):
|
|
|
56 |
return translated
|
57 |
return text
|
58 |
|
59 |
+
@spaces.GPU
|
60 |
def preprocess_image(image: Image.Image) -> Tuple[str, Image.Image]:
|
61 |
+
try:
|
62 |
+
trial_id = str(uuid.uuid4())
|
63 |
+
|
64 |
+
# ์ด๋ฏธ์ง๊ฐ ๋๋ฌด ์์ ๊ฒฝ์ฐ ํฌ๊ธฐ ์กฐ์
|
65 |
+
min_size = 64
|
66 |
+
if image.size[0] < min_size or image.size[1] < min_size:
|
67 |
+
ratio = min_size / min(image.size)
|
68 |
+
new_size = tuple(int(dim * ratio) for dim in image.size)
|
69 |
+
image = image.resize(new_size, Image.LANCZOS)
|
70 |
+
|
71 |
+
processed_image = pipeline.preprocess_image(image)
|
72 |
+
processed_image.save(f"{TMP_DIR}/{trial_id}.png")
|
73 |
+
return trial_id, processed_image
|
74 |
+
|
75 |
+
except Exception as e:
|
76 |
+
print(f"Error in preprocess_image: {str(e)}")
|
77 |
+
return None, None
|
78 |
|
79 |
def pack_state(gs: Gaussian, mesh: MeshExtractResult, trial_id: str) -> dict:
|
80 |
return {
|
|
|
93 |
'trial_id': trial_id,
|
94 |
}
|
95 |
|
|
|
96 |
def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
|
97 |
gs = Gaussian(
|
98 |
aabb=state['gaussian']['aabb'],
|
|
|
119 |
def image_to_3d(trial_id: str, seed: int, randomize_seed: bool, ss_guidance_strength: float,
|
120 |
ss_sampling_steps: int, slat_guidance_strength: float, slat_sampling_steps: int):
|
121 |
try:
|
|
|
|
|
122 |
if randomize_seed:
|
123 |
seed = np.random.randint(0, MAX_SEED)
|
124 |
|
125 |
input_image = Image.open(f"{TMP_DIR}/{trial_id}.png")
|
126 |
|
127 |
+
# GPU ์ค์
|
128 |
+
if torch.cuda.is_available():
|
129 |
+
pipeline.to("cuda")
|
130 |
+
pipeline.to(torch.float16)
|
131 |
+
|
132 |
+
with torch.no_grad():
|
133 |
+
outputs = pipeline.run(
|
134 |
+
input_image,
|
135 |
+
seed=seed,
|
136 |
+
formats=["gaussian", "mesh"],
|
137 |
+
preprocess_image=False,
|
138 |
+
sparse_structure_sampler_params={
|
139 |
+
"steps": ss_sampling_steps,
|
140 |
+
"cfg_strength": ss_guidance_strength,
|
141 |
+
},
|
142 |
+
slat_sampler_params={
|
143 |
+
"steps": slat_sampling_steps,
|
144 |
+
"cfg_strength": slat_guidance_strength,
|
145 |
+
}
|
146 |
+
)
|
147 |
|
|
|
148 |
video = render_utils.render_video(outputs['gaussian'][0], num_frames=120)['color']
|
149 |
video_geo = render_utils.render_video(outputs['mesh'][0], num_frames=120)['normal']
|
150 |
video = [np.concatenate([video[i], video_geo[i]], axis=1) for i in range(len(video))]
|
|
|
156 |
|
157 |
state = pack_state(outputs['gaussian'][0], outputs['mesh'][0], trial_id)
|
158 |
|
159 |
+
# CPU ๋ชจ๋๋ก ๋์๊ฐ๊ธฐ
|
160 |
+
pipeline.to("cpu")
|
161 |
+
|
162 |
return state, video_path
|
163 |
|
164 |
except Exception as e:
|
165 |
print(f"Error in image_to_3d: {str(e)}")
|
166 |
+
pipeline.to("cpu")
|
167 |
raise e
|
168 |
|
169 |
@spaces.GPU
|
170 |
def generate_image_from_text(prompt, height, width, guidance_scale, num_steps):
|
171 |
+
try:
|
172 |
+
# GPU ์ค์
|
173 |
+
if torch.cuda.is_available():
|
174 |
+
flux_pipe.to("cuda")
|
175 |
+
flux_pipe.to(torch.float16)
|
176 |
+
|
177 |
+
# ๊ธฐ๋ณธ ํ๋กฌํํธ๋ฅผ ์ถ๊ฐ
|
178 |
+
base_prompt = "wbgmsst, 3D, white background"
|
179 |
+
|
180 |
+
# ์ฌ์ฉ์ ํ๋กฌํํธ๋ฅผ ๋ฒ์ญ (ํ๊ตญ์ด์ธ ๊ฒฝ์ฐ)
|
181 |
+
translated_prompt = translate_if_korean(prompt)
|
182 |
+
|
183 |
+
# ์ต์ข
ํ๋กฌํํธ ์กฐํฉ
|
184 |
+
final_prompt = f"{translated_prompt}, {base_prompt}"
|
185 |
+
|
186 |
+
with torch.inference_mode():
|
187 |
+
image = flux_pipe(
|
188 |
+
prompt=[final_prompt],
|
189 |
+
height=height,
|
190 |
+
width=width,
|
191 |
+
guidance_scale=guidance_scale,
|
192 |
+
num_inference_steps=num_steps
|
193 |
+
).images[0]
|
194 |
+
|
195 |
+
# CPU ๋ชจ๋๋ก ๋์๊ฐ๊ธฐ
|
196 |
+
flux_pipe.to("cpu")
|
197 |
|
198 |
return image
|
199 |
+
|
200 |
+
except Exception as e:
|
201 |
+
print(f"Error in generate_image_from_text: {str(e)}")
|
202 |
+
flux_pipe.to("cpu")
|
203 |
+
raise e
|
204 |
|
205 |
@spaces.GPU
|
206 |
def extract_glb(state: dict, mesh_simplify: float, texture_size: int) -> Tuple[str, str]:
|
|
|
216 |
def deactivate_button() -> gr.Button:
|
217 |
return gr.Button(interactive=False)
|
218 |
|
|
|
219 |
css = """
|
220 |
footer {
|
221 |
visibility: hidden;
|
222 |
}
|
223 |
"""
|
224 |
|
225 |
+
# Gradio ์ธํฐํ์ด์ค ์ ์
|
226 |
with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
|
227 |
gr.Markdown("""
|
228 |
# Craft3D : 3D Asset Creation & Text-to-Image Generation
|
|
|
298 |
examples_per_page=64,
|
299 |
)
|
300 |
|
301 |
+
# Handlers
|
302 |
image_prompt.upload(
|
303 |
preprocess_image,
|
304 |
inputs=[image_prompt],
|
|
|
312 |
|
313 |
generate_btn.click(
|
314 |
image_to_3d,
|
315 |
+
inputs=[trial_id, seed, randomize_seed, ss_guidance_strength, ss_sampling_steps,
|
316 |
+
slat_guidance_strength, slat_sampling_steps],
|
317 |
outputs=[output_buf, video_output],
|
318 |
+
concurrency_limit=1
|
319 |
).then(
|
320 |
activate_button,
|
321 |
+
outputs=[extract_glb_btn]
|
|
|
|
|
|
|
|
|
|
|
322 |
)
|
323 |
|
324 |
extract_glb_btn.click(
|
325 |
extract_glb,
|
326 |
inputs=[output_buf, mesh_simplify, texture_size],
|
327 |
outputs=[model_output, download_glb],
|
328 |
+
concurrency_limit=1
|
329 |
).then(
|
330 |
activate_button,
|
331 |
+
outputs=[download_glb]
|
332 |
)
|
333 |
|
|
|
|
|
|
|
|
|
|
|
|
|
334 |
generate_txt2img_btn.click(
|
335 |
generate_image_from_text,
|
336 |
inputs=[text_prompt, txt2img_height, txt2img_width, guidance_scale, num_steps],
|
337 |
+
outputs=[txt2img_output],
|
338 |
+
concurrency_limit=1
|
339 |
)
|
340 |
|
341 |
if __name__ == "__main__":
|
342 |
+
# ๋ชจ๋ธ ์ด๊ธฐํ
|
|
|
|
|
|
|
|
|
343 |
if not initialize_models():
|
344 |
print("Failed to initialize models")
|
345 |
exit(1)
|
346 |
|
347 |
try:
|
348 |
# rembg ์ฌ์ ๋ก๋ ์๋
|
349 |
+
test_image = Image.fromarray(np.ones((256, 256, 3), dtype=np.uint8) * 255)
|
350 |
pipeline.preprocess_image(test_image)
|
351 |
except Exception as e:
|
352 |
print(f"Warning: Failed to preload rembg: {str(e)}")
|
353 |
|
354 |
# Gradio ์ฑ ์คํ
|
355 |
+
demo.queue(max_size=20).launch(
|
356 |
share=True,
|
357 |
+
max_threads=4,
|
358 |
+
show_error=True
|
359 |
)
|