Spaces:
Running
on
L40S
Running
on
L40S
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 |
os.environ['SPCONV_ALGO'] = 'native'
|
7 |
from typing import *
|
@@ -14,12 +13,13 @@ from PIL import Image
|
|
14 |
from trellis.pipelines import TrellisImageTo3DPipeline
|
15 |
from trellis.representations import Gaussian, MeshExtractResult
|
16 |
from trellis.utils import render_utils, postprocessing_utils
|
17 |
-
|
18 |
-
# 기존 import문 아래에 추가
|
19 |
from transformers import pipeline as translation_pipeline
|
20 |
from diffusers import FluxPipeline
|
21 |
|
22 |
-
|
|
|
|
|
|
|
23 |
def initialize_models():
|
24 |
global pipeline, translator, flux_pipe
|
25 |
|
@@ -35,30 +35,19 @@ def initialize_models():
|
|
35 |
flux_pipe.load_lora_weights("gokaygokay/Flux-Game-Assets-LoRA-v2")
|
36 |
flux_pipe.fuse_lora(lora_scale=1.0)
|
37 |
flux_pipe.to(device="cuda", dtype=torch.bfloat16)
|
38 |
-
|
39 |
-
MAX_SEED = np.iinfo(np.int32).max
|
40 |
-
TMP_DIR = "/tmp/Trellis-demo"
|
41 |
-
|
42 |
-
os.makedirs(TMP_DIR, exist_ok=True)
|
43 |
|
|
|
|
|
|
|
|
|
|
|
44 |
|
45 |
def preprocess_image(image: Image.Image) -> Tuple[str, Image.Image]:
|
46 |
-
"""
|
47 |
-
Preprocess the input image.
|
48 |
-
|
49 |
-
Args:
|
50 |
-
image (Image.Image): The input image.
|
51 |
-
|
52 |
-
Returns:
|
53 |
-
str: uuid of the trial.
|
54 |
-
Image.Image: The preprocessed image.
|
55 |
-
"""
|
56 |
trial_id = str(uuid.uuid4())
|
57 |
processed_image = pipeline.preprocess_image(image)
|
58 |
processed_image.save(f"{TMP_DIR}/{trial_id}.png")
|
59 |
return trial_id, processed_image
|
60 |
|
61 |
-
|
62 |
def pack_state(gs: Gaussian, mesh: MeshExtractResult, trial_id: str) -> dict:
|
63 |
return {
|
64 |
'gaussian': {
|
@@ -75,8 +64,8 @@ def pack_state(gs: Gaussian, mesh: MeshExtractResult, trial_id: str) -> dict:
|
|
75 |
},
|
76 |
'trial_id': trial_id,
|
77 |
}
|
78 |
-
|
79 |
-
|
80 |
def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
|
81 |
gs = Gaussian(
|
82 |
aabb=state['gaussian']['aabb'],
|
@@ -99,25 +88,8 @@ def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
|
|
99 |
|
100 |
return gs, mesh, state['trial_id']
|
101 |
|
102 |
-
|
103 |
@spaces.GPU
|
104 |
def image_to_3d(trial_id: str, seed: int, randomize_seed: bool, ss_guidance_strength: float, ss_sampling_steps: int, slat_guidance_strength: float, slat_sampling_steps: int) -> Tuple[dict, str]:
|
105 |
-
"""
|
106 |
-
Convert an image to a 3D model.
|
107 |
-
|
108 |
-
Args:
|
109 |
-
trial_id (str): The uuid of the trial.
|
110 |
-
seed (int): The random seed.
|
111 |
-
randomize_seed (bool): Whether to randomize the seed.
|
112 |
-
ss_guidance_strength (float): The guidance strength for sparse structure generation.
|
113 |
-
ss_sampling_steps (int): The number of sampling steps for sparse structure generation.
|
114 |
-
slat_guidance_strength (float): The guidance strength for structured latent generation.
|
115 |
-
slat_sampling_steps (int): The number of sampling steps for structured latent generation.
|
116 |
-
|
117 |
-
Returns:
|
118 |
-
dict: The information of the generated 3D model.
|
119 |
-
str: The path to the video of the 3D model.
|
120 |
-
"""
|
121 |
if randomize_seed:
|
122 |
seed = np.random.randint(0, MAX_SEED)
|
123 |
outputs = pipeline.run(
|
@@ -144,75 +116,98 @@ def image_to_3d(trial_id: str, seed: int, randomize_seed: bool, ss_guidance_stre
|
|
144 |
state = pack_state(outputs['gaussian'][0], outputs['mesh'][0], trial_id)
|
145 |
return state, video_path
|
146 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
147 |
|
148 |
@spaces.GPU
|
149 |
def extract_glb(state: dict, mesh_simplify: float, texture_size: int) -> Tuple[str, str]:
|
150 |
-
"""
|
151 |
-
Extract a GLB file from the 3D model.
|
152 |
-
|
153 |
-
Args:
|
154 |
-
state (dict): The state of the generated 3D model.
|
155 |
-
mesh_simplify (float): The mesh simplification factor.
|
156 |
-
texture_size (int): The texture resolution.
|
157 |
-
|
158 |
-
Returns:
|
159 |
-
str: The path to the extracted GLB file.
|
160 |
-
"""
|
161 |
gs, mesh, trial_id = unpack_state(state)
|
162 |
glb = postprocessing_utils.to_glb(gs, mesh, simplify=mesh_simplify, texture_size=texture_size, verbose=False)
|
163 |
glb_path = f"{TMP_DIR}/{trial_id}.glb"
|
164 |
glb.export(glb_path)
|
165 |
return glb_path, glb_path
|
166 |
|
167 |
-
|
168 |
def activate_button() -> gr.Button:
|
169 |
return gr.Button(interactive=True)
|
170 |
|
171 |
-
|
172 |
def deactivate_button() -> gr.Button:
|
173 |
return gr.Button(interactive=False)
|
174 |
|
175 |
|
176 |
with gr.Blocks() as demo:
|
177 |
gr.Markdown("""
|
178 |
-
|
179 |
-
* Upload an image and click "Generate" to create a 3D asset. If the image has alpha channel, it be used as the mask. Otherwise, we use `rembg` to remove the background.
|
180 |
-
* If you find the generated 3D asset satisfactory, click "Extract GLB" to extract the GLB file and download it.
|
181 |
""")
|
182 |
|
183 |
-
with gr.
|
184 |
-
with gr.
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
212 |
trial_id = gr.Textbox(visible=False)
|
213 |
output_buf = gr.State()
|
214 |
|
215 |
-
# Example images
|
216 |
with gr.Row():
|
217 |
examples = gr.Examples(
|
218 |
examples=[
|
@@ -226,12 +221,13 @@ with gr.Blocks() as demo:
|
|
226 |
examples_per_page=64,
|
227 |
)
|
228 |
|
229 |
-
|
230 |
image_prompt.upload(
|
231 |
preprocess_image,
|
232 |
inputs=[image_prompt],
|
233 |
outputs=[trial_id, image_prompt],
|
234 |
)
|
|
|
235 |
image_prompt.clear(
|
236 |
lambda: '',
|
237 |
outputs=[trial_id],
|
@@ -264,14 +260,19 @@ with gr.Blocks() as demo:
|
|
264 |
deactivate_button,
|
265 |
outputs=[download_glb],
|
266 |
)
|
267 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
268 |
|
269 |
# Launch the Gradio app
|
270 |
if __name__ == "__main__":
|
271 |
-
|
272 |
-
pipeline.cuda()
|
273 |
try:
|
274 |
pipeline.preprocess_image(Image.fromarray(np.zeros((512, 512, 3), dtype=np.uint8))) # Preload rembg
|
275 |
except:
|
276 |
pass
|
277 |
-
demo.launch()
|
|
|
1 |
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 *
|
|
|
13 |
from trellis.pipelines import TrellisImageTo3DPipeline
|
14 |
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"
|
21 |
+
os.makedirs(TMP_DIR, exist_ok=True)
|
22 |
+
|
23 |
def initialize_models():
|
24 |
global pipeline, translator, flux_pipe
|
25 |
|
|
|
35 |
flux_pipe.load_lora_weights("gokaygokay/Flux-Game-Assets-LoRA-v2")
|
36 |
flux_pipe.fuse_lora(lora_scale=1.0)
|
37 |
flux_pipe.to(device="cuda", dtype=torch.bfloat16)
|
|
|
|
|
|
|
|
|
|
|
38 |
|
39 |
+
def translate_if_korean(text):
|
40 |
+
if any(ord('가') <= ord(char) <= ord('힣') for char in text):
|
41 |
+
translated = translator(text)[0]['translation_text']
|
42 |
+
return translated
|
43 |
+
return text
|
44 |
|
45 |
def preprocess_image(image: Image.Image) -> Tuple[str, Image.Image]:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
trial_id = str(uuid.uuid4())
|
47 |
processed_image = pipeline.preprocess_image(image)
|
48 |
processed_image.save(f"{TMP_DIR}/{trial_id}.png")
|
49 |
return trial_id, processed_image
|
50 |
|
|
|
51 |
def pack_state(gs: Gaussian, mesh: MeshExtractResult, trial_id: str) -> dict:
|
52 |
return {
|
53 |
'gaussian': {
|
|
|
64 |
},
|
65 |
'trial_id': trial_id,
|
66 |
}
|
67 |
+
|
68 |
+
|
69 |
def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
|
70 |
gs = Gaussian(
|
71 |
aabb=state['gaussian']['aabb'],
|
|
|
88 |
|
89 |
return gs, mesh, state['trial_id']
|
90 |
|
|
|
91 |
@spaces.GPU
|
92 |
def image_to_3d(trial_id: str, seed: int, randomize_seed: bool, ss_guidance_strength: float, ss_sampling_steps: int, slat_guidance_strength: float, slat_sampling_steps: int) -> Tuple[dict, str]:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
93 |
if randomize_seed:
|
94 |
seed = np.random.randint(0, MAX_SEED)
|
95 |
outputs = pipeline.run(
|
|
|
116 |
state = pack_state(outputs['gaussian'][0], outputs['mesh'][0], trial_id)
|
117 |
return state, video_path
|
118 |
|
119 |
+
@spaces.GPU
|
120 |
+
def generate_image_from_text(prompt, height, width, guidance_scale, num_steps):
|
121 |
+
translated_prompt = translate_if_korean(prompt)
|
122 |
+
|
123 |
+
with torch.inference_mode():
|
124 |
+
image = flux_pipe(
|
125 |
+
prompt=[translated_prompt],
|
126 |
+
height=height,
|
127 |
+
width=width,
|
128 |
+
guidance_scale=guidance_scale,
|
129 |
+
num_inference_steps=num_steps
|
130 |
+
).images[0]
|
131 |
+
|
132 |
+
return image
|
133 |
|
134 |
@spaces.GPU
|
135 |
def extract_glb(state: dict, mesh_simplify: float, texture_size: int) -> Tuple[str, str]:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
gs, mesh, trial_id = unpack_state(state)
|
137 |
glb = postprocessing_utils.to_glb(gs, mesh, simplify=mesh_simplify, texture_size=texture_size, verbose=False)
|
138 |
glb_path = f"{TMP_DIR}/{trial_id}.glb"
|
139 |
glb.export(glb_path)
|
140 |
return glb_path, glb_path
|
141 |
|
|
|
142 |
def activate_button() -> gr.Button:
|
143 |
return gr.Button(interactive=True)
|
144 |
|
|
|
145 |
def deactivate_button() -> gr.Button:
|
146 |
return gr.Button(interactive=False)
|
147 |
|
148 |
|
149 |
with gr.Blocks() as demo:
|
150 |
gr.Markdown("""
|
151 |
+
# 3D Asset Creation & Text-to-Image Generation
|
|
|
|
|
152 |
""")
|
153 |
|
154 |
+
with gr.Tabs():
|
155 |
+
with gr.TabItem("Image to 3D"):
|
156 |
+
with gr.Row():
|
157 |
+
with gr.Column():
|
158 |
+
image_prompt = gr.Image(label="Image Prompt", image_mode="RGBA", type="pil", height=300)
|
159 |
+
|
160 |
+
with gr.Accordion(label="Generation Settings", open=False):
|
161 |
+
seed = gr.Slider(0, MAX_SEED, label="Seed", value=0, step=1)
|
162 |
+
randomize_seed = gr.Checkbox(label="Randomize Seed", value=True)
|
163 |
+
gr.Markdown("Stage 1: Sparse Structure Generation")
|
164 |
+
with gr.Row():
|
165 |
+
ss_guidance_strength = gr.Slider(0.0, 10.0, label="Guidance Strength", value=7.5, step=0.1)
|
166 |
+
ss_sampling_steps = gr.Slider(1, 50, label="Sampling Steps", value=12, step=1)
|
167 |
+
gr.Markdown("Stage 2: Structured Latent Generation")
|
168 |
+
with gr.Row():
|
169 |
+
slat_guidance_strength = gr.Slider(0.0, 10.0, label="Guidance Strength", value=3.0, step=0.1)
|
170 |
+
slat_sampling_steps = gr.Slider(1, 50, label="Sampling Steps", value=12, step=1)
|
171 |
+
|
172 |
+
generate_btn = gr.Button("Generate")
|
173 |
+
|
174 |
+
with gr.Accordion(label="GLB Extraction Settings", open=False):
|
175 |
+
mesh_simplify = gr.Slider(0.9, 0.98, label="Simplify", value=0.95, step=0.01)
|
176 |
+
texture_size = gr.Slider(512, 2048, label="Texture Size", value=1024, step=512)
|
177 |
+
|
178 |
+
extract_glb_btn = gr.Button("Extract GLB", interactive=False)
|
179 |
+
|
180 |
+
with gr.Column():
|
181 |
+
video_output = gr.Video(label="Generated 3D Asset", autoplay=True, loop=True, height=300)
|
182 |
+
model_output = LitModel3D(label="Extracted GLB", exposure=20.0, height=300)
|
183 |
+
download_glb = gr.DownloadButton(label="Download GLB", interactive=False)
|
184 |
+
|
185 |
+
with gr.TabItem("Text to Image"):
|
186 |
+
with gr.Row():
|
187 |
+
with gr.Column():
|
188 |
+
text_prompt = gr.Textbox(
|
189 |
+
label="Text Prompt",
|
190 |
+
placeholder="Enter your image description...",
|
191 |
+
lines=3
|
192 |
+
)
|
193 |
+
|
194 |
+
with gr.Row():
|
195 |
+
txt2img_height = gr.Slider(256, 1024, value=512, step=64, label="Height")
|
196 |
+
txt2img_width = gr.Slider(256, 1024, value=512, step=64, label="Width")
|
197 |
+
|
198 |
+
with gr.Row():
|
199 |
+
guidance_scale = gr.Slider(1.0, 20.0, value=7.5, label="Guidance Scale")
|
200 |
+
num_steps = gr.Slider(1, 50, value=20, label="Number of Steps")
|
201 |
+
|
202 |
+
generate_txt2img_btn = gr.Button("Generate Image")
|
203 |
+
|
204 |
+
with gr.Column():
|
205 |
+
txt2img_output = gr.Image(label="Generated Image")
|
206 |
+
|
207 |
trial_id = gr.Textbox(visible=False)
|
208 |
output_buf = gr.State()
|
209 |
|
210 |
+
# Example images
|
211 |
with gr.Row():
|
212 |
examples = gr.Examples(
|
213 |
examples=[
|
|
|
221 |
examples_per_page=64,
|
222 |
)
|
223 |
|
224 |
+
# Handlers
|
225 |
image_prompt.upload(
|
226 |
preprocess_image,
|
227 |
inputs=[image_prompt],
|
228 |
outputs=[trial_id, image_prompt],
|
229 |
)
|
230 |
+
|
231 |
image_prompt.clear(
|
232 |
lambda: '',
|
233 |
outputs=[trial_id],
|
|
|
260 |
deactivate_button,
|
261 |
outputs=[download_glb],
|
262 |
)
|
263 |
+
|
264 |
+
# Text to Image 핸들러
|
265 |
+
generate_txt2img_btn.click(
|
266 |
+
generate_image_from_text,
|
267 |
+
inputs=[text_prompt, txt2img_height, txt2img_width, guidance_scale, num_steps],
|
268 |
+
outputs=[txt2img_output]
|
269 |
+
)
|
270 |
|
271 |
# Launch the Gradio app
|
272 |
if __name__ == "__main__":
|
273 |
+
initialize_models() # 모든 모델 초기화
|
|
|
274 |
try:
|
275 |
pipeline.preprocess_image(Image.fromarray(np.zeros((512, 512, 3), dtype=np.uint8))) # Preload rembg
|
276 |
except:
|
277 |
pass
|
278 |
+
demo.launch()
|