JeffreyXiang
commited on
Commit
•
b7b00e2
1
Parent(s):
cd41f5f
Add multiimage and gaussian
Browse files- app.py +171 -38
- assets/example_multi_image/character_1.png +0 -0
- assets/example_multi_image/character_2.png +0 -0
- assets/example_multi_image/character_3.png +0 -0
- assets/example_multi_image/mushroom_1.png +0 -0
- assets/example_multi_image/mushroom_2.png +0 -0
- assets/example_multi_image/mushroom_3.png +0 -0
- assets/example_multi_image/orangeguy_1.png +0 -0
- assets/example_multi_image/orangeguy_2.png +0 -0
- assets/example_multi_image/orangeguy_3.png +0 -0
- assets/example_multi_image/popmart_1.png +0 -0
- assets/example_multi_image/popmart_2.png +0 -0
- assets/example_multi_image/popmart_3.png +0 -0
- assets/example_multi_image/rabbit_1.png +0 -0
- assets/example_multi_image/rabbit_2.png +0 -0
- assets/example_multi_image/rabbit_3.png +0 -0
- assets/example_multi_image/tiger_1.png +0 -0
- assets/example_multi_image/tiger_2.png +0 -0
- assets/example_multi_image/tiger_3.png +0 -0
- assets/example_multi_image/yoimiya_1.png +0 -0
- assets/example_multi_image/yoimiya_2.png +0 -0
- assets/example_multi_image/yoimiya_3.png +0 -0
- trellis/pipelines/trellis_image_to_3d.py +93 -0
- trellis/representations/gaussian/gaussian_model.py +18 -3
- trellis/utils/postprocessing_utils.py +130 -1
app.py
CHANGED
@@ -9,7 +9,6 @@ from typing import *
|
|
9 |
import torch
|
10 |
import numpy as np
|
11 |
import imageio
|
12 |
-
import uuid
|
13 |
from easydict import EasyDict as edict
|
14 |
from PIL import Image
|
15 |
from trellis.pipelines import TrellisImageTo3DPipeline
|
@@ -24,17 +23,15 @@ os.makedirs(TMP_DIR, exist_ok=True)
|
|
24 |
|
25 |
def start_session(req: gr.Request):
|
26 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
27 |
-
print(f'Creating user directory: {user_dir}')
|
28 |
os.makedirs(user_dir, exist_ok=True)
|
29 |
|
30 |
|
31 |
def end_session(req: gr.Request):
|
32 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
33 |
-
print(f'Removing user directory: {user_dir}')
|
34 |
shutil.rmtree(user_dir)
|
35 |
|
36 |
|
37 |
-
def preprocess_image(image: Image.Image) ->
|
38 |
"""
|
39 |
Preprocess the input image.
|
40 |
|
@@ -42,14 +39,28 @@ def preprocess_image(image: Image.Image) -> Tuple[str, Image.Image]:
|
|
42 |
image (Image.Image): The input image.
|
43 |
|
44 |
Returns:
|
45 |
-
str: uuid of the trial.
|
46 |
Image.Image: The preprocessed image.
|
47 |
"""
|
48 |
processed_image = pipeline.preprocess_image(image)
|
49 |
return processed_image
|
50 |
|
51 |
|
52 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
return {
|
54 |
'gaussian': {
|
55 |
**gs.init_params,
|
@@ -63,7 +74,6 @@ def pack_state(gs: Gaussian, mesh: MeshExtractResult, trial_id: str) -> dict:
|
|
63 |
'vertices': mesh.vertices.cpu().numpy(),
|
64 |
'faces': mesh.faces.cpu().numpy(),
|
65 |
},
|
66 |
-
'trial_id': trial_id,
|
67 |
}
|
68 |
|
69 |
|
@@ -87,7 +97,7 @@ def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
|
|
87 |
faces=torch.tensor(state['mesh']['faces'], device='cuda'),
|
88 |
)
|
89 |
|
90 |
-
return gs, mesh
|
91 |
|
92 |
|
93 |
def get_seed(randomize_seed: bool, seed: int) -> int:
|
@@ -100,11 +110,14 @@ def get_seed(randomize_seed: bool, seed: int) -> int:
|
|
100 |
@spaces.GPU
|
101 |
def image_to_3d(
|
102 |
image: Image.Image,
|
|
|
|
|
103 |
seed: int,
|
104 |
ss_guidance_strength: float,
|
105 |
ss_sampling_steps: int,
|
106 |
slat_guidance_strength: float,
|
107 |
slat_sampling_steps: int,
|
|
|
108 |
req: gr.Request,
|
109 |
) -> Tuple[dict, str]:
|
110 |
"""
|
@@ -112,43 +125,62 @@ def image_to_3d(
|
|
112 |
|
113 |
Args:
|
114 |
image (Image.Image): The input image.
|
|
|
|
|
115 |
seed (int): The random seed.
|
116 |
ss_guidance_strength (float): The guidance strength for sparse structure generation.
|
117 |
ss_sampling_steps (int): The number of sampling steps for sparse structure generation.
|
118 |
slat_guidance_strength (float): The guidance strength for structured latent generation.
|
119 |
slat_sampling_steps (int): The number of sampling steps for structured latent generation.
|
|
|
120 |
|
121 |
Returns:
|
122 |
dict: The information of the generated 3D model.
|
123 |
str: The path to the video of the 3D model.
|
124 |
"""
|
125 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
140 |
video = render_utils.render_video(outputs['gaussian'][0], num_frames=120)['color']
|
141 |
video_geo = render_utils.render_video(outputs['mesh'][0], num_frames=120)['normal']
|
142 |
video = [np.concatenate([video[i], video_geo[i]], axis=1) for i in range(len(video))]
|
143 |
-
|
144 |
-
video_path = os.path.join(user_dir, f"{trial_id}.mp4")
|
145 |
imageio.mimsave(video_path, video, fps=15)
|
146 |
-
state = pack_state(outputs['gaussian'][0], outputs['mesh'][0]
|
147 |
torch.cuda.empty_cache()
|
148 |
return state, video_path
|
149 |
|
150 |
|
151 |
-
@spaces.GPU
|
152 |
def extract_glb(
|
153 |
state: dict,
|
154 |
mesh_simplify: float,
|
@@ -167,24 +199,83 @@ def extract_glb(
|
|
167 |
str: The path to the extracted GLB file.
|
168 |
"""
|
169 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
170 |
-
gs, mesh
|
171 |
glb = postprocessing_utils.to_glb(gs, mesh, simplify=mesh_simplify, texture_size=texture_size, verbose=False)
|
172 |
-
glb_path = os.path.join(user_dir,
|
173 |
glb.export(glb_path)
|
174 |
torch.cuda.empty_cache()
|
175 |
return glb_path, glb_path
|
176 |
|
177 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
178 |
with gr.Blocks(delete_cache=(600, 600)) as demo:
|
179 |
gr.Markdown("""
|
180 |
## Image to 3D Asset with [TRELLIS](https://trellis3d.github.io/)
|
181 |
* 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.
|
182 |
* If you find the generated 3D asset satisfactory, click "Extract GLB" to extract the GLB file and download it.
|
|
|
|
|
183 |
""")
|
184 |
|
185 |
with gr.Row():
|
186 |
with gr.Column():
|
187 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
188 |
|
189 |
with gr.Accordion(label="Generation Settings", open=False):
|
190 |
seed = gr.Slider(0, MAX_SEED, label="Seed", value=0, step=1)
|
@@ -197,6 +288,7 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
197 |
with gr.Row():
|
198 |
slat_guidance_strength = gr.Slider(0.0, 10.0, label="Guidance Strength", value=3.0, step=0.1)
|
199 |
slat_sampling_steps = gr.Slider(1, 50, label="Sampling Steps", value=12, step=1)
|
|
|
200 |
|
201 |
generate_btn = gr.Button("Generate")
|
202 |
|
@@ -204,17 +296,26 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
204 |
mesh_simplify = gr.Slider(0.9, 0.98, label="Simplify", value=0.95, step=0.01)
|
205 |
texture_size = gr.Slider(512, 2048, label="Texture Size", value=1024, step=512)
|
206 |
|
207 |
-
|
|
|
|
|
|
|
|
|
|
|
208 |
|
209 |
with gr.Column():
|
210 |
video_output = gr.Video(label="Generated 3D Asset", autoplay=True, loop=True, height=300)
|
211 |
-
model_output = LitModel3D(label="Extracted GLB", exposure=
|
212 |
-
download_glb = gr.DownloadButton(label="Download GLB", interactive=False)
|
213 |
|
|
|
|
|
|
|
|
|
|
|
214 |
output_buf = gr.State()
|
215 |
|
216 |
# Example images at the bottom of the page
|
217 |
-
with gr.Row():
|
218 |
examples = gr.Examples(
|
219 |
examples=[
|
220 |
f'assets/example_image/{image}'
|
@@ -226,16 +327,39 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
226 |
run_on_click=True,
|
227 |
examples_per_page=64,
|
228 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
229 |
|
230 |
# Handlers
|
231 |
demo.load(start_session)
|
232 |
demo.unload(end_session)
|
233 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
234 |
image_prompt.upload(
|
235 |
preprocess_image,
|
236 |
inputs=[image_prompt],
|
237 |
outputs=[image_prompt],
|
238 |
)
|
|
|
|
|
|
|
|
|
|
|
239 |
|
240 |
generate_btn.click(
|
241 |
get_seed,
|
@@ -243,16 +367,16 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
243 |
outputs=[seed],
|
244 |
).then(
|
245 |
image_to_3d,
|
246 |
-
inputs=[image_prompt, seed, ss_guidance_strength, ss_sampling_steps, slat_guidance_strength, slat_sampling_steps],
|
247 |
outputs=[output_buf, video_output],
|
248 |
).then(
|
249 |
-
lambda: gr.Button(interactive=True),
|
250 |
-
outputs=[extract_glb_btn],
|
251 |
)
|
252 |
|
253 |
video_output.clear(
|
254 |
-
lambda: gr.Button(interactive=False),
|
255 |
-
outputs=[extract_glb_btn],
|
256 |
)
|
257 |
|
258 |
extract_glb_btn.click(
|
@@ -263,6 +387,15 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
263 |
lambda: gr.Button(interactive=True),
|
264 |
outputs=[download_glb],
|
265 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
266 |
|
267 |
model_output.clear(
|
268 |
lambda: gr.Button(interactive=False),
|
|
|
9 |
import torch
|
10 |
import numpy as np
|
11 |
import imageio
|
|
|
12 |
from easydict import EasyDict as edict
|
13 |
from PIL import Image
|
14 |
from trellis.pipelines import TrellisImageTo3DPipeline
|
|
|
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_image(image: Image.Image) -> Image.Image:
|
35 |
"""
|
36 |
Preprocess the input image.
|
37 |
|
|
|
39 |
image (Image.Image): The input image.
|
40 |
|
41 |
Returns:
|
|
|
42 |
Image.Image: The preprocessed image.
|
43 |
"""
|
44 |
processed_image = pipeline.preprocess_image(image)
|
45 |
return processed_image
|
46 |
|
47 |
|
48 |
+
def preprocess_images(images: List[Tuple[Image.Image, str]]) -> List[Image.Image]:
|
49 |
+
"""
|
50 |
+
Preprocess a list of input images.
|
51 |
+
|
52 |
+
Args:
|
53 |
+
images (List[Tuple[Image.Image, str]]): The input images.
|
54 |
+
|
55 |
+
Returns:
|
56 |
+
List[Image.Image]: The preprocessed images.
|
57 |
+
"""
|
58 |
+
images = [image[0] for image in images]
|
59 |
+
processed_images = [pipeline.preprocess_image(image) for image in images]
|
60 |
+
return processed_images
|
61 |
+
|
62 |
+
|
63 |
+
def pack_state(gs: Gaussian, mesh: MeshExtractResult) -> dict:
|
64 |
return {
|
65 |
'gaussian': {
|
66 |
**gs.init_params,
|
|
|
74 |
'vertices': mesh.vertices.cpu().numpy(),
|
75 |
'faces': mesh.faces.cpu().numpy(),
|
76 |
},
|
|
|
77 |
}
|
78 |
|
79 |
|
|
|
97 |
faces=torch.tensor(state['mesh']['faces'], device='cuda'),
|
98 |
)
|
99 |
|
100 |
+
return gs, mesh
|
101 |
|
102 |
|
103 |
def get_seed(randomize_seed: bool, seed: int) -> int:
|
|
|
110 |
@spaces.GPU
|
111 |
def image_to_3d(
|
112 |
image: Image.Image,
|
113 |
+
multiimages: List[Tuple[Image.Image, str]],
|
114 |
+
is_multiimage: bool,
|
115 |
seed: int,
|
116 |
ss_guidance_strength: float,
|
117 |
ss_sampling_steps: int,
|
118 |
slat_guidance_strength: float,
|
119 |
slat_sampling_steps: int,
|
120 |
+
multiimage_algo: Literal["multidiffusion", "stochastic"],
|
121 |
req: gr.Request,
|
122 |
) -> Tuple[dict, str]:
|
123 |
"""
|
|
|
125 |
|
126 |
Args:
|
127 |
image (Image.Image): The input image.
|
128 |
+
multiimages (List[Tuple[Image.Image, str]]): The input images in multi-image mode.
|
129 |
+
is_multiimage (bool): Whether is in multi-image mode.
|
130 |
seed (int): The random seed.
|
131 |
ss_guidance_strength (float): The guidance strength for sparse structure generation.
|
132 |
ss_sampling_steps (int): The number of sampling steps for sparse structure generation.
|
133 |
slat_guidance_strength (float): The guidance strength for structured latent generation.
|
134 |
slat_sampling_steps (int): The number of sampling steps for structured latent generation.
|
135 |
+
multiimage_algo (Literal["multidiffusion", "stochastic"]): The algorithm for multi-image generation.
|
136 |
|
137 |
Returns:
|
138 |
dict: The information of the generated 3D model.
|
139 |
str: The path to the video of the 3D model.
|
140 |
"""
|
141 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
142 |
+
if not is_multiimage:
|
143 |
+
outputs = pipeline.run(
|
144 |
+
image,
|
145 |
+
seed=seed,
|
146 |
+
formats=["gaussian", "mesh"],
|
147 |
+
preprocess_image=False,
|
148 |
+
sparse_structure_sampler_params={
|
149 |
+
"steps": ss_sampling_steps,
|
150 |
+
"cfg_strength": ss_guidance_strength,
|
151 |
+
},
|
152 |
+
slat_sampler_params={
|
153 |
+
"steps": slat_sampling_steps,
|
154 |
+
"cfg_strength": slat_guidance_strength,
|
155 |
+
},
|
156 |
+
)
|
157 |
+
else:
|
158 |
+
outputs = pipeline.run_multi_image(
|
159 |
+
[image[0] for image in multiimages],
|
160 |
+
seed=seed,
|
161 |
+
formats=["gaussian", "mesh"],
|
162 |
+
preprocess_image=False,
|
163 |
+
sparse_structure_sampler_params={
|
164 |
+
"steps": ss_sampling_steps,
|
165 |
+
"cfg_strength": ss_guidance_strength,
|
166 |
+
},
|
167 |
+
slat_sampler_params={
|
168 |
+
"steps": slat_sampling_steps,
|
169 |
+
"cfg_strength": slat_guidance_strength,
|
170 |
+
},
|
171 |
+
mode=multiimage_algo,
|
172 |
+
)
|
173 |
video = render_utils.render_video(outputs['gaussian'][0], num_frames=120)['color']
|
174 |
video_geo = render_utils.render_video(outputs['mesh'][0], num_frames=120)['normal']
|
175 |
video = [np.concatenate([video[i], video_geo[i]], axis=1) for i in range(len(video))]
|
176 |
+
video_path = os.path.join(user_dir, 'sample.mp4')
|
|
|
177 |
imageio.mimsave(video_path, video, fps=15)
|
178 |
+
state = pack_state(outputs['gaussian'][0], outputs['mesh'][0])
|
179 |
torch.cuda.empty_cache()
|
180 |
return state, video_path
|
181 |
|
182 |
|
183 |
+
@spaces.GPU(duration=90)
|
184 |
def extract_glb(
|
185 |
state: dict,
|
186 |
mesh_simplify: float,
|
|
|
199 |
str: The path to the extracted GLB file.
|
200 |
"""
|
201 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
202 |
+
gs, mesh = unpack_state(state)
|
203 |
glb = postprocessing_utils.to_glb(gs, mesh, simplify=mesh_simplify, texture_size=texture_size, verbose=False)
|
204 |
+
glb_path = os.path.join(user_dir, 'sample.glb')
|
205 |
glb.export(glb_path)
|
206 |
torch.cuda.empty_cache()
|
207 |
return glb_path, glb_path
|
208 |
|
209 |
|
210 |
+
@spaces.GPU
|
211 |
+
def extract_gaussian(state: dict, req: gr.Request) -> Tuple[str, str]:
|
212 |
+
"""
|
213 |
+
Extract a Gaussian file from the 3D model.
|
214 |
+
|
215 |
+
Args:
|
216 |
+
state (dict): The state of the generated 3D model.
|
217 |
+
|
218 |
+
Returns:
|
219 |
+
str: The path to the extracted Gaussian file.
|
220 |
+
"""
|
221 |
+
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
222 |
+
gs, _ = unpack_state(state)
|
223 |
+
gaussian_path = os.path.join(user_dir, 'sample.ply')
|
224 |
+
gs.save_ply(gaussian_path)
|
225 |
+
torch.cuda.empty_cache()
|
226 |
+
return gaussian_path, gaussian_path
|
227 |
+
|
228 |
+
|
229 |
+
def prepare_multi_example() -> List[Image.Image]:
|
230 |
+
multi_case = list(set([i.split('_')[0] for i in os.listdir("assets/example_multi_image")]))
|
231 |
+
images = []
|
232 |
+
for case in multi_case:
|
233 |
+
_images = []
|
234 |
+
for i in range(1, 4):
|
235 |
+
img = Image.open(f'assets/example_multi_image/{case}_{i}.png')
|
236 |
+
W, H = img.size
|
237 |
+
img = img.resize((int(W / H * 512), 512))
|
238 |
+
_images.append(np.array(img))
|
239 |
+
images.append(Image.fromarray(np.concatenate(_images, axis=1)))
|
240 |
+
return images
|
241 |
+
|
242 |
+
|
243 |
+
def split_image(image: Image.Image) -> List[Image.Image]:
|
244 |
+
"""
|
245 |
+
Split an image into multiple views.
|
246 |
+
"""
|
247 |
+
image = np.array(image)
|
248 |
+
alpha = image[..., 3]
|
249 |
+
alpha = np.any(alpha>0, axis=0)
|
250 |
+
start_pos = np.where(~alpha[:-1] & alpha[1:])[0].tolist()
|
251 |
+
end_pos = np.where(alpha[:-1] & ~alpha[1:])[0].tolist()
|
252 |
+
images = []
|
253 |
+
for s, e in zip(start_pos, end_pos):
|
254 |
+
images.append(Image.fromarray(image[:, s:e+1]))
|
255 |
+
return [preprocess_image(image) for image in images]
|
256 |
+
|
257 |
+
|
258 |
with gr.Blocks(delete_cache=(600, 600)) as demo:
|
259 |
gr.Markdown("""
|
260 |
## Image to 3D Asset with [TRELLIS](https://trellis3d.github.io/)
|
261 |
* 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.
|
262 |
* If you find the generated 3D asset satisfactory, click "Extract GLB" to extract the GLB file and download it.
|
263 |
+
|
264 |
+
✨New: 1) Experimental multi-image support. 2) Gaussian file extraction.
|
265 |
""")
|
266 |
|
267 |
with gr.Row():
|
268 |
with gr.Column():
|
269 |
+
with gr.Tabs() as input_tabs:
|
270 |
+
with gr.Tab(label="Single Image", id=0) as single_image_input_tab:
|
271 |
+
image_prompt = gr.Image(label="Image Prompt", format="png", image_mode="RGBA", type="pil", height=300)
|
272 |
+
with gr.Tab(label="Multiple Images", id=1) as multiimage_input_tab:
|
273 |
+
multiimage_prompt = gr.Gallery(label="Image Prompt", format="png", type="pil", height=300, columns=3)
|
274 |
+
gr.Markdown("""
|
275 |
+
Input different views of the object in separate images.
|
276 |
+
|
277 |
+
*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.*
|
278 |
+
""")
|
279 |
|
280 |
with gr.Accordion(label="Generation Settings", open=False):
|
281 |
seed = gr.Slider(0, MAX_SEED, label="Seed", value=0, step=1)
|
|
|
288 |
with gr.Row():
|
289 |
slat_guidance_strength = gr.Slider(0.0, 10.0, label="Guidance Strength", value=3.0, step=0.1)
|
290 |
slat_sampling_steps = gr.Slider(1, 50, label="Sampling Steps", value=12, step=1)
|
291 |
+
multiimage_algo = gr.Radio(["stochastic", "multidiffusion"], label="Multi-image Algorithm", value="stochastic")
|
292 |
|
293 |
generate_btn = gr.Button("Generate")
|
294 |
|
|
|
296 |
mesh_simplify = gr.Slider(0.9, 0.98, label="Simplify", value=0.95, step=0.01)
|
297 |
texture_size = gr.Slider(512, 2048, label="Texture Size", value=1024, step=512)
|
298 |
|
299 |
+
with gr.Row():
|
300 |
+
extract_glb_btn = gr.Button("Extract GLB", interactive=False)
|
301 |
+
extract_gs_btn = gr.Button("Extract Gaussian", interactive=False)
|
302 |
+
gr.Markdown("""
|
303 |
+
*NOTE: Gaussian file can be very large (~50MB), it will take a while to display and download.*
|
304 |
+
""")
|
305 |
|
306 |
with gr.Column():
|
307 |
video_output = gr.Video(label="Generated 3D Asset", autoplay=True, loop=True, height=300)
|
308 |
+
model_output = LitModel3D(label="Extracted GLB/Gaussian", exposure=10.0, height=300)
|
|
|
309 |
|
310 |
+
with gr.Row():
|
311 |
+
download_glb = gr.DownloadButton(label="Download GLB", interactive=False)
|
312 |
+
download_gs = gr.DownloadButton(label="Download Gaussian", interactive=False)
|
313 |
+
|
314 |
+
is_multiimage = gr.State(False)
|
315 |
output_buf = gr.State()
|
316 |
|
317 |
# Example images at the bottom of the page
|
318 |
+
with gr.Row() as single_image_example:
|
319 |
examples = gr.Examples(
|
320 |
examples=[
|
321 |
f'assets/example_image/{image}'
|
|
|
327 |
run_on_click=True,
|
328 |
examples_per_page=64,
|
329 |
)
|
330 |
+
with gr.Row(visible=False) as multiimage_example:
|
331 |
+
examples_multi = gr.Examples(
|
332 |
+
examples=prepare_multi_example(),
|
333 |
+
inputs=[image_prompt],
|
334 |
+
fn=split_image,
|
335 |
+
outputs=[multiimage_prompt],
|
336 |
+
run_on_click=True,
|
337 |
+
examples_per_page=8,
|
338 |
+
)
|
339 |
|
340 |
# Handlers
|
341 |
demo.load(start_session)
|
342 |
demo.unload(end_session)
|
343 |
|
344 |
+
single_image_input_tab.select(
|
345 |
+
lambda: tuple([False, gr.Row.update(visible=True), gr.Row.update(visible=False)]),
|
346 |
+
outputs=[is_multiimage, single_image_example, multiimage_example]
|
347 |
+
)
|
348 |
+
multiimage_input_tab.select(
|
349 |
+
lambda: tuple([True, gr.Row.update(visible=False), gr.Row.update(visible=True)]),
|
350 |
+
outputs=[is_multiimage, single_image_example, multiimage_example]
|
351 |
+
)
|
352 |
+
|
353 |
image_prompt.upload(
|
354 |
preprocess_image,
|
355 |
inputs=[image_prompt],
|
356 |
outputs=[image_prompt],
|
357 |
)
|
358 |
+
multiimage_prompt.upload(
|
359 |
+
preprocess_images,
|
360 |
+
inputs=[multiimage_prompt],
|
361 |
+
outputs=[multiimage_prompt],
|
362 |
+
)
|
363 |
|
364 |
generate_btn.click(
|
365 |
get_seed,
|
|
|
367 |
outputs=[seed],
|
368 |
).then(
|
369 |
image_to_3d,
|
370 |
+
inputs=[image_prompt, multiimage_prompt, is_multiimage, seed, ss_guidance_strength, ss_sampling_steps, slat_guidance_strength, slat_sampling_steps, multiimage_algo],
|
371 |
outputs=[output_buf, video_output],
|
372 |
).then(
|
373 |
+
lambda: tuple([gr.Button(interactive=True), gr.Button(interactive=True)]),
|
374 |
+
outputs=[extract_glb_btn, extract_gs_btn],
|
375 |
)
|
376 |
|
377 |
video_output.clear(
|
378 |
+
lambda: tuple([gr.Button(interactive=False), gr.Button(interactive=False)]),
|
379 |
+
outputs=[extract_glb_btn, extract_gs_btn],
|
380 |
)
|
381 |
|
382 |
extract_glb_btn.click(
|
|
|
387 |
lambda: gr.Button(interactive=True),
|
388 |
outputs=[download_glb],
|
389 |
)
|
390 |
+
|
391 |
+
extract_gs_btn.click(
|
392 |
+
extract_gaussian,
|
393 |
+
inputs=[output_buf],
|
394 |
+
outputs=[model_output, download_gs],
|
395 |
+
).then(
|
396 |
+
lambda: gr.Button(interactive=True),
|
397 |
+
outputs=[download_gs],
|
398 |
+
)
|
399 |
|
400 |
model_output.clear(
|
401 |
lambda: gr.Button(interactive=False),
|
assets/example_multi_image/character_1.png
ADDED
assets/example_multi_image/character_2.png
ADDED
assets/example_multi_image/character_3.png
ADDED
assets/example_multi_image/mushroom_1.png
ADDED
assets/example_multi_image/mushroom_2.png
ADDED
assets/example_multi_image/mushroom_3.png
ADDED
assets/example_multi_image/orangeguy_1.png
ADDED
assets/example_multi_image/orangeguy_2.png
ADDED
assets/example_multi_image/orangeguy_3.png
ADDED
assets/example_multi_image/popmart_1.png
ADDED
assets/example_multi_image/popmart_2.png
ADDED
assets/example_multi_image/popmart_3.png
ADDED
assets/example_multi_image/rabbit_1.png
ADDED
assets/example_multi_image/rabbit_2.png
ADDED
assets/example_multi_image/rabbit_3.png
ADDED
assets/example_multi_image/tiger_1.png
ADDED
assets/example_multi_image/tiger_2.png
ADDED
assets/example_multi_image/tiger_3.png
ADDED
assets/example_multi_image/yoimiya_1.png
ADDED
assets/example_multi_image/yoimiya_2.png
ADDED
assets/example_multi_image/yoimiya_3.png
ADDED
trellis/pipelines/trellis_image_to_3d.py
CHANGED
@@ -1,4 +1,5 @@
|
|
1 |
from typing import *
|
|
|
2 |
import torch
|
3 |
import torch.nn as nn
|
4 |
import torch.nn.functional as F
|
@@ -281,3 +282,95 @@ class TrellisImageTo3DPipeline(Pipeline):
|
|
281 |
coords = self.sample_sparse_structure(cond, num_samples, sparse_structure_sampler_params)
|
282 |
slat = self.sample_slat(cond, coords, slat_sampler_params)
|
283 |
return self.decode_slat(slat, formats)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
from typing import *
|
2 |
+
from contextlib import contextmanager
|
3 |
import torch
|
4 |
import torch.nn as nn
|
5 |
import torch.nn.functional as F
|
|
|
282 |
coords = self.sample_sparse_structure(cond, num_samples, sparse_structure_sampler_params)
|
283 |
slat = self.sample_slat(cond, coords, slat_sampler_params)
|
284 |
return self.decode_slat(slat, formats)
|
285 |
+
|
286 |
+
@contextmanager
|
287 |
+
def inject_sampler_multi_image(
|
288 |
+
self,
|
289 |
+
sampler_name: str,
|
290 |
+
num_images: int,
|
291 |
+
num_steps: int,
|
292 |
+
mode: Literal['stochastic', 'multidiffusion'] = 'stochastic',
|
293 |
+
):
|
294 |
+
"""
|
295 |
+
Inject a sampler with multiple images as condition.
|
296 |
+
|
297 |
+
Args:
|
298 |
+
sampler_name (str): The name of the sampler to inject.
|
299 |
+
num_images (int): The number of images to condition on.
|
300 |
+
num_steps (int): The number of steps to run the sampler for.
|
301 |
+
"""
|
302 |
+
sampler = getattr(self, sampler_name)
|
303 |
+
setattr(sampler, f'_old_inference_model', sampler._inference_model)
|
304 |
+
|
305 |
+
if mode == 'stochastic':
|
306 |
+
if num_images > num_steps:
|
307 |
+
print(f"\033[93mWarning: number of conditioning images is greater than number of steps for {sampler_name}. "
|
308 |
+
"This may lead to performance degradation.\033[0m")
|
309 |
+
|
310 |
+
cond_indices = (np.arange(num_steps) % num_images).tolist()
|
311 |
+
def _new_inference_model(self, model, x_t, t, cond, **kwargs):
|
312 |
+
cond_idx = cond_indices.pop(0)
|
313 |
+
cond_i = cond[cond_idx:cond_idx+1]
|
314 |
+
return self._old_inference_model(model, x_t, t, cond=cond_i, **kwargs)
|
315 |
+
|
316 |
+
elif mode =='multidiffusion':
|
317 |
+
from .samplers import FlowEulerSampler
|
318 |
+
def _new_inference_model(self, model, x_t, t, cond, neg_cond, cfg_strength, cfg_interval, **kwargs):
|
319 |
+
if cfg_interval[0] <= t <= cfg_interval[1]:
|
320 |
+
preds = []
|
321 |
+
for i in range(len(cond)):
|
322 |
+
preds.append(FlowEulerSampler._inference_model(self, model, x_t, t, cond[i:i+1], **kwargs))
|
323 |
+
pred = sum(preds) / len(preds)
|
324 |
+
neg_pred = FlowEulerSampler._inference_model(self, model, x_t, t, neg_cond, **kwargs)
|
325 |
+
return (1 + cfg_strength) * pred - cfg_strength * neg_pred
|
326 |
+
else:
|
327 |
+
preds = []
|
328 |
+
for i in range(len(cond)):
|
329 |
+
preds.append(FlowEulerSampler._inference_model(self, model, x_t, t, cond[i:i+1], **kwargs))
|
330 |
+
pred = sum(preds) / len(preds)
|
331 |
+
return pred
|
332 |
+
|
333 |
+
else:
|
334 |
+
raise ValueError(f"Unsupported mode: {mode}")
|
335 |
+
|
336 |
+
sampler._inference_model = _new_inference_model.__get__(sampler, type(sampler))
|
337 |
+
|
338 |
+
yield
|
339 |
+
|
340 |
+
sampler._inference_model = sampler._old_inference_model
|
341 |
+
delattr(sampler, f'_old_inference_model')
|
342 |
+
|
343 |
+
@torch.no_grad()
|
344 |
+
def run_multi_image(
|
345 |
+
self,
|
346 |
+
images: List[Image.Image],
|
347 |
+
num_samples: int = 1,
|
348 |
+
seed: int = 42,
|
349 |
+
sparse_structure_sampler_params: dict = {},
|
350 |
+
slat_sampler_params: dict = {},
|
351 |
+
formats: List[str] = ['mesh', 'gaussian', 'radiance_field'],
|
352 |
+
preprocess_image: bool = True,
|
353 |
+
mode: Literal['stochastic', 'multidiffusion'] = 'stochastic',
|
354 |
+
) -> dict:
|
355 |
+
"""
|
356 |
+
Run the pipeline with multiple images as condition
|
357 |
+
|
358 |
+
Args:
|
359 |
+
images (List[Image.Image]): The multi-view images of the assets
|
360 |
+
num_samples (int): The number of samples to generate.
|
361 |
+
sparse_structure_sampler_params (dict): Additional parameters for the sparse structure sampler.
|
362 |
+
slat_sampler_params (dict): Additional parameters for the structured latent sampler.
|
363 |
+
preprocess_image (bool): Whether to preprocess the image.
|
364 |
+
"""
|
365 |
+
if preprocess_image:
|
366 |
+
images = [self.preprocess_image(image) for image in images]
|
367 |
+
cond = self.get_cond(images)
|
368 |
+
cond['neg_cond'] = cond['neg_cond'][:1]
|
369 |
+
torch.manual_seed(seed)
|
370 |
+
ss_steps = {**self.sparse_structure_sampler_params, **sparse_structure_sampler_params}.get('steps')
|
371 |
+
with self.inject_sampler_multi_image('sparse_structure_sampler', len(images), ss_steps, mode=mode):
|
372 |
+
coords = self.sample_sparse_structure(cond, num_samples, sparse_structure_sampler_params)
|
373 |
+
slat_steps = {**self.slat_sampler_params, **slat_sampler_params}.get('steps')
|
374 |
+
with self.inject_sampler_multi_image('slat_sampler', len(images), slat_steps, mode=mode):
|
375 |
+
slat = self.sample_slat(cond, coords, slat_sampler_params)
|
376 |
+
return self.decode_slat(slat, formats)
|
trellis/representations/gaussian/gaussian_model.py
CHANGED
@@ -2,6 +2,7 @@ import torch
|
|
2 |
import numpy as np
|
3 |
from plyfile import PlyData, PlyElement
|
4 |
from .general_utils import inverse_sigmoid, strip_symmetric, build_scaling_rotation
|
|
|
5 |
|
6 |
|
7 |
class Gaussian:
|
@@ -120,14 +121,21 @@ class Gaussian:
|
|
120 |
for i in range(self._rotation.shape[1]):
|
121 |
l.append('rot_{}'.format(i))
|
122 |
return l
|
123 |
-
|
124 |
-
def save_ply(self, path):
|
125 |
xyz = self.get_xyz.detach().cpu().numpy()
|
126 |
normals = np.zeros_like(xyz)
|
127 |
f_dc = self._features_dc.detach().transpose(1, 2).flatten(start_dim=1).contiguous().cpu().numpy()
|
128 |
opacities = inverse_sigmoid(self.get_opacity).detach().cpu().numpy()
|
129 |
scale = torch.log(self.get_scaling).detach().cpu().numpy()
|
130 |
rotation = (self._rotation + self.rots_bias[None, :]).detach().cpu().numpy()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
131 |
|
132 |
dtype_full = [(attribute, 'f4') for attribute in self.construct_list_of_attributes()]
|
133 |
|
@@ -137,7 +145,7 @@ class Gaussian:
|
|
137 |
el = PlyElement.describe(elements, 'vertex')
|
138 |
PlyData([el]).write(path)
|
139 |
|
140 |
-
def load_ply(self, path):
|
141 |
plydata = PlyData.read(path)
|
142 |
|
143 |
xyz = np.stack((np.asarray(plydata.elements[0]["x"]),
|
@@ -172,6 +180,13 @@ class Gaussian:
|
|
172 |
for idx, attr_name in enumerate(rot_names):
|
173 |
rots[:, idx] = np.asarray(plydata.elements[0][attr_name])
|
174 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
175 |
# convert to actual gaussian attributes
|
176 |
xyz = torch.tensor(xyz, dtype=torch.float, device=self.device)
|
177 |
features_dc = torch.tensor(features_dc, dtype=torch.float, device=self.device).transpose(1, 2).contiguous()
|
|
|
2 |
import numpy as np
|
3 |
from plyfile import PlyData, PlyElement
|
4 |
from .general_utils import inverse_sigmoid, strip_symmetric, build_scaling_rotation
|
5 |
+
import utils3d
|
6 |
|
7 |
|
8 |
class Gaussian:
|
|
|
121 |
for i in range(self._rotation.shape[1]):
|
122 |
l.append('rot_{}'.format(i))
|
123 |
return l
|
124 |
+
|
125 |
+
def save_ply(self, path, transform=[[1, 0, 0], [0, 0, -1], [0, 1, 0]]):
|
126 |
xyz = self.get_xyz.detach().cpu().numpy()
|
127 |
normals = np.zeros_like(xyz)
|
128 |
f_dc = self._features_dc.detach().transpose(1, 2).flatten(start_dim=1).contiguous().cpu().numpy()
|
129 |
opacities = inverse_sigmoid(self.get_opacity).detach().cpu().numpy()
|
130 |
scale = torch.log(self.get_scaling).detach().cpu().numpy()
|
131 |
rotation = (self._rotation + self.rots_bias[None, :]).detach().cpu().numpy()
|
132 |
+
|
133 |
+
if transform is not None:
|
134 |
+
transform = np.array(transform)
|
135 |
+
xyz = np.matmul(xyz, transform.T)
|
136 |
+
rotation = utils3d.numpy.quaternion_to_matrix(rotation)
|
137 |
+
rotation = np.matmul(transform, rotation)
|
138 |
+
rotation = utils3d.numpy.matrix_to_quaternion(rotation)
|
139 |
|
140 |
dtype_full = [(attribute, 'f4') for attribute in self.construct_list_of_attributes()]
|
141 |
|
|
|
145 |
el = PlyElement.describe(elements, 'vertex')
|
146 |
PlyData([el]).write(path)
|
147 |
|
148 |
+
def load_ply(self, path, transform=[[1, 0, 0], [0, 0, -1], [0, 1, 0]]):
|
149 |
plydata = PlyData.read(path)
|
150 |
|
151 |
xyz = np.stack((np.asarray(plydata.elements[0]["x"]),
|
|
|
180 |
for idx, attr_name in enumerate(rot_names):
|
181 |
rots[:, idx] = np.asarray(plydata.elements[0][attr_name])
|
182 |
|
183 |
+
if transform is not None:
|
184 |
+
transform = np.array(transform)
|
185 |
+
xyz = np.matmul(xyz, transform)
|
186 |
+
rotation = utils3d.numpy.quaternion_to_matrix(rotation)
|
187 |
+
rotation = np.matmul(rotation, transform)
|
188 |
+
rotation = utils3d.numpy.matrix_to_quaternion(rotation)
|
189 |
+
|
190 |
# convert to actual gaussian attributes
|
191 |
xyz = torch.tensor(xyz, dtype=torch.float, device=self.device)
|
192 |
features_dc = torch.tensor(features_dc, dtype=torch.float, device=self.device).transpose(1, 2).contiguous()
|
trellis/utils/postprocessing_utils.py
CHANGED
@@ -14,6 +14,7 @@ import cv2
|
|
14 |
from PIL import Image
|
15 |
from .random_utils import sphere_hammersley_sequence
|
16 |
from .render_utils import render_multiview
|
|
|
17 |
from ..representations import Strivec, Gaussian, MeshExtractResult
|
18 |
|
19 |
|
@@ -454,5 +455,133 @@ def to_glb(
|
|
454 |
|
455 |
# rotate mesh (from z-up to y-up)
|
456 |
vertices = vertices @ np.array([[1, 0, 0], [0, 0, -1], [0, 1, 0]])
|
457 |
-
|
|
|
|
|
|
|
|
|
|
|
458 |
return mesh
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
from PIL import Image
|
15 |
from .random_utils import sphere_hammersley_sequence
|
16 |
from .render_utils import render_multiview
|
17 |
+
from ..renderers import GaussianRenderer
|
18 |
from ..representations import Strivec, Gaussian, MeshExtractResult
|
19 |
|
20 |
|
|
|
455 |
|
456 |
# rotate mesh (from z-up to y-up)
|
457 |
vertices = vertices @ np.array([[1, 0, 0], [0, 0, -1], [0, 1, 0]])
|
458 |
+
material = trimesh.visual.material.PBRMaterial(
|
459 |
+
roughnessFactor=1.0,
|
460 |
+
baseColorTexture=texture,
|
461 |
+
baseColorFactor=np.array([255, 255, 255, 255], dtype=np.uint8)
|
462 |
+
)
|
463 |
+
mesh = trimesh.Trimesh(vertices, faces, visual=trimesh.visual.TextureVisuals(uv=uvs, material=material))
|
464 |
return mesh
|
465 |
+
|
466 |
+
|
467 |
+
def simplify_gs(
|
468 |
+
gs: Gaussian,
|
469 |
+
simplify: float = 0.95,
|
470 |
+
verbose: bool = True,
|
471 |
+
):
|
472 |
+
"""
|
473 |
+
Simplify 3D Gaussians
|
474 |
+
NOTE: this function is not used in the current implementation for the unsatisfactory performance.
|
475 |
+
|
476 |
+
Args:
|
477 |
+
gs (Gaussian): 3D Gaussian.
|
478 |
+
simplify (float): Ratio of Gaussians to remove in simplification.
|
479 |
+
"""
|
480 |
+
if simplify <= 0:
|
481 |
+
return gs
|
482 |
+
|
483 |
+
# simplify
|
484 |
+
observations, extrinsics, intrinsics = render_multiview(gs, resolution=1024, nviews=100)
|
485 |
+
observations = [torch.tensor(obs / 255.0).float().cuda().permute(2, 0, 1) for obs in observations]
|
486 |
+
|
487 |
+
# Following https://arxiv.org/pdf/2411.06019
|
488 |
+
renderer = GaussianRenderer({
|
489 |
+
"resolution": 1024,
|
490 |
+
"near": 0.8,
|
491 |
+
"far": 1.6,
|
492 |
+
"ssaa": 1,
|
493 |
+
"bg_color": (0,0,0),
|
494 |
+
})
|
495 |
+
new_gs = Gaussian(**gs.init_params)
|
496 |
+
new_gs._features_dc = gs._features_dc.clone()
|
497 |
+
new_gs._features_rest = gs._features_rest.clone() if gs._features_rest is not None else None
|
498 |
+
new_gs._opacity = torch.nn.Parameter(gs._opacity.clone())
|
499 |
+
new_gs._rotation = torch.nn.Parameter(gs._rotation.clone())
|
500 |
+
new_gs._scaling = torch.nn.Parameter(gs._scaling.clone())
|
501 |
+
new_gs._xyz = torch.nn.Parameter(gs._xyz.clone())
|
502 |
+
|
503 |
+
start_lr = [1e-4, 1e-3, 5e-3, 0.025]
|
504 |
+
end_lr = [1e-6, 1e-5, 5e-5, 0.00025]
|
505 |
+
optimizer = torch.optim.Adam([
|
506 |
+
{"params": new_gs._xyz, "lr": start_lr[0]},
|
507 |
+
{"params": new_gs._rotation, "lr": start_lr[1]},
|
508 |
+
{"params": new_gs._scaling, "lr": start_lr[2]},
|
509 |
+
{"params": new_gs._opacity, "lr": start_lr[3]},
|
510 |
+
], lr=start_lr[0])
|
511 |
+
|
512 |
+
def exp_anealing(optimizer, step, total_steps, start_lr, end_lr):
|
513 |
+
return start_lr * (end_lr / start_lr) ** (step / total_steps)
|
514 |
+
|
515 |
+
def cosine_anealing(optimizer, step, total_steps, start_lr, end_lr):
|
516 |
+
return end_lr + 0.5 * (start_lr - end_lr) * (1 + np.cos(np.pi * step / total_steps))
|
517 |
+
|
518 |
+
_zeta = new_gs.get_opacity.clone().detach().squeeze()
|
519 |
+
_lambda = torch.zeros_like(_zeta)
|
520 |
+
_delta = 1e-7
|
521 |
+
_interval = 10
|
522 |
+
num_target = int((1 - simplify) * _zeta.shape[0])
|
523 |
+
|
524 |
+
with tqdm(total=2500, disable=not verbose, desc='Simplifying Gaussian') as pbar:
|
525 |
+
for i in range(2500):
|
526 |
+
# prune
|
527 |
+
if i % 100 == 0:
|
528 |
+
mask = new_gs.get_opacity.squeeze() > 0.05
|
529 |
+
mask = torch.nonzero(mask).squeeze()
|
530 |
+
new_gs._xyz = torch.nn.Parameter(new_gs._xyz[mask])
|
531 |
+
new_gs._rotation = torch.nn.Parameter(new_gs._rotation[mask])
|
532 |
+
new_gs._scaling = torch.nn.Parameter(new_gs._scaling[mask])
|
533 |
+
new_gs._opacity = torch.nn.Parameter(new_gs._opacity[mask])
|
534 |
+
new_gs._features_dc = new_gs._features_dc[mask]
|
535 |
+
new_gs._features_rest = new_gs._features_rest[mask] if new_gs._features_rest is not None else None
|
536 |
+
_zeta = _zeta[mask]
|
537 |
+
_lambda = _lambda[mask]
|
538 |
+
# update optimizer state
|
539 |
+
for param_group, new_param in zip(optimizer.param_groups, [new_gs._xyz, new_gs._rotation, new_gs._scaling, new_gs._opacity]):
|
540 |
+
stored_state = optimizer.state[param_group['params'][0]]
|
541 |
+
if 'exp_avg' in stored_state:
|
542 |
+
stored_state['exp_avg'] = stored_state['exp_avg'][mask]
|
543 |
+
stored_state['exp_avg_sq'] = stored_state['exp_avg_sq'][mask]
|
544 |
+
del optimizer.state[param_group['params'][0]]
|
545 |
+
param_group['params'][0] = new_param
|
546 |
+
optimizer.state[param_group['params'][0]] = stored_state
|
547 |
+
|
548 |
+
opacity = new_gs.get_opacity.squeeze()
|
549 |
+
|
550 |
+
# sparisfy
|
551 |
+
if i % _interval == 0:
|
552 |
+
_zeta = _lambda + opacity.detach()
|
553 |
+
if opacity.shape[0] > num_target:
|
554 |
+
index = _zeta.topk(num_target)[1]
|
555 |
+
_m = torch.ones_like(_zeta, dtype=torch.bool)
|
556 |
+
_m[index] = 0
|
557 |
+
_zeta[_m] = 0
|
558 |
+
_lambda = _lambda + opacity.detach() - _zeta
|
559 |
+
|
560 |
+
# sample a random view
|
561 |
+
view_idx = np.random.randint(len(observations))
|
562 |
+
observation = observations[view_idx]
|
563 |
+
extrinsic = extrinsics[view_idx]
|
564 |
+
intrinsic = intrinsics[view_idx]
|
565 |
+
|
566 |
+
color = renderer.render(new_gs, extrinsic, intrinsic)['color']
|
567 |
+
rgb_loss = torch.nn.functional.l1_loss(color, observation)
|
568 |
+
loss = rgb_loss + \
|
569 |
+
_delta * torch.sum(torch.pow(_lambda + opacity - _zeta, 2))
|
570 |
+
|
571 |
+
optimizer.zero_grad()
|
572 |
+
loss.backward()
|
573 |
+
optimizer.step()
|
574 |
+
|
575 |
+
# update lr
|
576 |
+
for j in range(len(optimizer.param_groups)):
|
577 |
+
optimizer.param_groups[j]['lr'] = cosine_anealing(optimizer, i, 2500, start_lr[j], end_lr[j])
|
578 |
+
|
579 |
+
pbar.set_postfix({'loss': rgb_loss.item(), 'num': opacity.shape[0], 'lambda': _lambda.mean().item()})
|
580 |
+
pbar.update()
|
581 |
+
|
582 |
+
new_gs._xyz = new_gs._xyz.data
|
583 |
+
new_gs._rotation = new_gs._rotation.data
|
584 |
+
new_gs._scaling = new_gs._scaling.data
|
585 |
+
new_gs._opacity = new_gs._opacity.data
|
586 |
+
|
587 |
+
return new_gs
|