File size: 8,769 Bytes
f108aa8
 
 
c5402af
 
f108aa8
c5402af
 
f108aa8
c5402af
d8c8548
 
 
c5402af
fcfad21
f108aa8
c5402af
f108aa8
 
1c1ff82
 
 
 
 
 
 
 
 
 
9177085
1c1ff82
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0ae4746
1c1ff82
 
 
 
 
 
 
 
 
9177085
 
 
 
 
 
 
 
ac4feba
ce940a6
1c1ff82
 
 
 
d8c8548
f108aa8
15066eb
 
 
 
 
 
f108aa8
 
8dd9712
f108aa8
 
 
 
 
 
 
 
 
 
 
 
e1a0427
 
 
 
f108aa8
 
8dd9712
d37df73
b3f7f27
8dd9712
 
 
 
 
2d034f7
 
 
f108aa8
 
 
1e9e245
f108aa8
8dd9712
f108aa8
e20f68b
f108aa8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6a82d6a
 
e1a0427
 
63bf201
e1a0427
f108aa8
 
 
b3f7f27
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
72559fa
b3f7f27
 
ac4feba
 
 
 
 
 
 
b3f7f27
 
 
ac4feba
 
 
 
f96485a
ac4feba
 
 
 
 
b3f7f27
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f108aa8
 
 
ba33fe9
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
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
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
import spaces
import gradio as gr
import os
import sys
from glob import glob
import time
from typing import Any, Union

import numpy as np
import torch

print(f'torch version:{torch.__version__}')

import trimesh
import glob
from huggingface_hub import snapshot_download
from PIL import Image
from accelerate.utils import set_seed

import subprocess
import importlib, site, sys

# Re-discover all .pth/.egg-link files
for sitedir in site.getsitepackages():
    site.addsitedir(sitedir)

# Clear caches so importlib will pick up new modules
importlib.invalidate_caches()

def sh(cmd): subprocess.check_call(cmd, shell=True)

def install_cuda_toolkit():
    CUDA_TOOLKIT_URL = "https://developer.download.nvidia.com/compute/cuda/12.6.0/local_installers/cuda_12.6.0_560.28.03_linux.run"
    CUDA_TOOLKIT_FILE = "/tmp/%s" % os.path.basename(CUDA_TOOLKIT_URL)
    subprocess.check_call(["wget", "-q", CUDA_TOOLKIT_URL, "-O", CUDA_TOOLKIT_FILE])
    subprocess.check_call(["chmod", "+x", CUDA_TOOLKIT_FILE])
    subprocess.check_call([CUDA_TOOLKIT_FILE, "--silent", "--toolkit"])

    os.environ["CUDA_HOME"] = "/usr/local/cuda"
    os.environ["PATH"] = "%s/bin:%s" % (os.environ["CUDA_HOME"], os.environ["PATH"])
    os.environ["LD_LIBRARY_PATH"] = "%s/lib:%s" % (
        os.environ["CUDA_HOME"],
        "" if "LD_LIBRARY_PATH" not in os.environ else os.environ["LD_LIBRARY_PATH"],
    )
    # add for compiler header lookup
    os.environ["CPATH"] = f"{os.environ['CUDA_HOME']}/include" + (
        f":{os.environ['CPATH']}" if "CPATH" in os.environ else ""
    )
    # Fix: arch_list[-1] += '+PTX'; IndexError: list index out of range
    os.environ["TORCH_CUDA_ARCH_LIST"] = "8.9;9.0"
    print("==> finished installation")

print("installing cuda toolkit")
install_cuda_toolkit()
print("finished")

header_path = "/usr/local/cuda/include/cuda_runtime.h"
print(f"{header_path} exists:", os.path.exists(header_path))

def sh(cmd_list, extra_env=None):
    env = os.environ.copy()
    if extra_env:
        env.update(extra_env)
    subprocess.check_call(cmd_list, env=env)

# install with FORCE_CUDA=1
sh(["pip", "install", "diso"], {"FORCE_CUDA": "1"})
# sh(["pip", "install", "torch-cluster", "-f", "https://data.pyg.org/whl/torch-2.7.0+126.html"])



# tell Python to re-scan site-packages now that the egg-link exists
import importlib, site; site.addsitedir(site.getsitepackages()[0]); importlib.invalidate_caches()


from src.utils.data_utils import get_colored_mesh_composition, scene_to_parts, load_surfaces
from src.utils.render_utils import render_views_around_mesh, render_normal_views_around_mesh, make_grid_for_images_or_videos, export_renderings
from src.pipelines.pipeline_partcrafter import PartCrafterPipeline
from src.utils.image_utils import prepare_image
from src.models.briarmbg import BriaRMBG

# Constants
MAX_NUM_PARTS = 16
DEVICE = "cuda" 
DTYPE = torch.float16

# Download and initialize models
partcrafter_weights_dir = "pretrained_weights/PartCrafter"
rmbg_weights_dir = "pretrained_weights/RMBG-1.4"
snapshot_download(repo_id="wgsxm/PartCrafter", local_dir=partcrafter_weights_dir)
snapshot_download(repo_id="briaai/RMBG-1.4", local_dir=rmbg_weights_dir)

rmbg_net = BriaRMBG.from_pretrained(rmbg_weights_dir).to(DEVICE)
rmbg_net.eval()
pipe: PartCrafterPipeline = PartCrafterPipeline.from_pretrained(partcrafter_weights_dir).to(DEVICE, DTYPE)

def first_file_from_dir(directory, ext):
    files = glob.glob(os.path.join(directory, f"*.{ext}"))
    return sorted(files)[0] if files else None

@spaces.GPU()
@torch.no_grad()
def run_triposg(image_path: str,
                num_parts: int = 1,
                seed: int = 0,
                num_tokens: int = 1024,
                num_inference_steps: int = 50,
                guidance_scale: float = 7.0,
                use_flash_decoder: bool = False,
                rmbg: bool = True):

    max_num_expanded_coords = 1e9
    
    """
    Generate 3D part meshes from an input image.
    """

    if rmbg:
        img_pil = prepare_image(image_path, bg_color=np.array([1.0, 1.0, 1.0]), rmbg_net=rmbg_net)
    else:
        img_pil = Image.open(image_path)

    set_seed(seed)
    start_time = time.time()
    outputs = pipe(
        image=[img_pil] * num_parts,
        attention_kwargs={"num_parts": num_parts},
        num_tokens=num_tokens,
        generator=torch.Generator(device=pipe.device).manual_seed(seed),
        num_inference_steps=num_inference_steps,
        guidance_scale=guidance_scale,
        max_num_expanded_coords=max_num_expanded_coords,
        use_flash_decoder=use_flash_decoder,
    ).meshes
    duration = time.time() - start_time
    print(f"Generation time: {duration:.2f}s")

    # Ensure no None outputs
    for i, mesh in enumerate(outputs):
        if mesh is None:
            outputs[i] = trimesh.Trimesh(vertices=[[0,0,0]], faces=[[0,0,0]])

    # Merge and color
    merged = get_colored_mesh_composition(outputs)

    # Export meshes and return results
    timestamp = time.strftime("%Y%m%d_%H%M%S")
    export_dir = os.path.join("results", timestamp)
    os.makedirs(export_dir, exist_ok=True)
    for idx, mesh in enumerate(outputs):
        mesh.export(os.path.join(export_dir, f"part_{idx:02}.glb"))

    glb_path = os.path.join(export_dir, "object.glb")
    merged.export(glb_path)

    mesh_file = first_file_from_dir(export_dir, "glb")
    
    return mesh_file, export_dir

# Gradio Interface
def build_demo():
    css = """
        #col-container {
            margin: 0 auto;
            max-width: 1024px;
        }
        """
    theme = gr.themes.Ocean()
    
    with gr.Blocks(css=css, theme=theme) as demo:
    
        with gr.Column(elem_id="col-container"):

            gr.Markdown(
            """ # PartCrafter – Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers
    
            • Source: [Github](https://github.com/wgsxm/PartCrafter)  
            • HF Space by : [@alexandernasa](https://twitter.com/alexandernasa/)  """
            )
            with gr.Row():
                with gr.Column(scale=1):
                    input_image = gr.Image(type="filepath", label="Input Image")
                    num_parts = gr.Slider(1, MAX_NUM_PARTS, value=4, step=1, label="Number of Parts")
                    run_button = gr.Button("Generate 3D Parts", variant="primary")
                    
                    with gr.Accordion("Advanced Settings", open=False):
                        seed = gr.Number(value=0, label="Random Seed", precision=0)
                        num_tokens = gr.Slider(256, 2048, value=1024, step=64, label="Num Tokens")
                        num_steps = gr.Slider(1, 100, value=50, step=1, label="Inference Steps")
                        guidance = gr.Slider(1.0, 20.0, value=7.0, step=0.1, label="Guidance Scale")
                        flash_decoder = gr.Checkbox(value=False, label="Use Flash Decoder")
                        remove_bg = gr.Checkbox(value=True, label="Remove Background (RMBG)")

                with gr.Column(scale=1):
                    gr.HTML(
                        """
                        <p style="opacity: 0.6; font-style: italic;">
                          This might take a few seconds to load the 3D model
                        </p>
                        """
                    )
                    output_model = gr.Model3D(label="Merged 3D Object")
                    output_dir = gr.Textbox(label="Export Directory")
                    examples = gr.Examples(
                        
                        examples=[
                            [
                                "assets/images/np5_b81f29e567ea4db48014f89c9079e403.png", 
                                5,
                            ], 
                            [
                                "assets/images/np10_cc486e491a2c499f9fd2aad2b02c6ccb.png", 
                                10,
                            ], 
                            [
                                "assets/images/np4_7bd5d25aa77b4fb18e780d7a4c97d342.png", 
                                4,
                            ], 
                        ],
                        inputs=[input_image, num_parts],
                        outputs=[output_model, output_dir],
                        fn=run_triposg,
                        cache_examples=True,
                    )
    
            run_button.click(fn=run_triposg,
                             inputs=[input_image, num_parts, seed, num_tokens, num_steps,
                                     guidance, flash_decoder, remove_bg],
                             outputs=[output_model, output_dir])
        return demo

if __name__ == "__main__":
    demo = build_demo()
    demo.launch()