Spaces:
Running
on
Zero
Running
on
Zero
bennyguo
commited on
Commit
·
2c25d73
1
Parent(s):
c60b074
initial demo release
Browse files- app.py +260 -4
- requirements.txt +16 -0
app.py
CHANGED
@@ -1,7 +1,263 @@
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
|
|
|
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
import os
|
3 |
+
import sys
|
4 |
+
import subprocess
|
5 |
+
from huggingface_hub import snapshot_download, HfFolder
|
6 |
+
import random # Import random for seed generation
|
7 |
|
8 |
+
# --- Repo Setup ---
|
9 |
+
DEFAULT_REPO_DIR = "./TripoSG-repo" # Directory to clone into if not using local path
|
10 |
+
REPO_GIT_URL = "github.com/VAST-AI-Research/TripoSG.git" # Base URL without schema/token
|
11 |
+
BRANCH = "scribble"
|
12 |
|
13 |
+
code_source_path = None
|
14 |
+
|
15 |
+
# Option 1: Use local path if TRIPOSG_CODE_PATH env var is set
|
16 |
+
local_code_path = os.environ.get("TRIPOSG_CODE_PATH")
|
17 |
+
if local_code_path:
|
18 |
+
print(f"Attempting to use local code path specified by TRIPOSG_CODE_PATH: {local_code_path}")
|
19 |
+
# Basic check: does it exist and seem like a git repo (has .git)?
|
20 |
+
if os.path.isdir(local_code_path) and os.path.isdir(os.path.join(local_code_path, ".git")):
|
21 |
+
code_source_path = os.path.abspath(local_code_path)
|
22 |
+
print(f"Using local TripoSG code directory: {code_source_path}")
|
23 |
+
# You might want to add a check here to verify the branch is correct, e.g.:
|
24 |
+
# try:
|
25 |
+
# current_branch = subprocess.run(["git", "rev-parse", "--abbrev-ref", "HEAD"], cwd=code_source_path, check=True, capture_output=True, text=True).stdout.strip()
|
26 |
+
# if current_branch != BRANCH:
|
27 |
+
# print(f"Warning: Local repo is on branch '{current_branch}', expected '{BRANCH}'. Attempting checkout...")
|
28 |
+
# subprocess.run(["git", "checkout", BRANCH], cwd=code_source_path, check=True)
|
29 |
+
# except Exception as e:
|
30 |
+
# print(f"Warning: Could not verify or checkout branch '{BRANCH}' in {code_source_path}: {e}")
|
31 |
+
else:
|
32 |
+
print(f"Warning: TRIPOSG_CODE_PATH '{local_code_path}' not found or not a valid git repository directory. Falling back to cloning.")
|
33 |
+
|
34 |
+
# Option 2: Clone from GitHub (if local path not used or invalid)
|
35 |
+
if not code_source_path:
|
36 |
+
repo_url_to_clone = f"https://{REPO_GIT_URL}"
|
37 |
+
github_token = os.environ.get("GITHUB_TOKEN")
|
38 |
+
if github_token:
|
39 |
+
print("Using GITHUB_TOKEN for repository cloning.")
|
40 |
+
repo_url_to_clone = f"https://{github_token}@{REPO_GIT_URL}"
|
41 |
+
else:
|
42 |
+
print("No GITHUB_TOKEN found. Using public HTTPS for cloning.")
|
43 |
+
|
44 |
+
repo_target_dir = os.path.abspath(DEFAULT_REPO_DIR)
|
45 |
+
if not os.path.exists(repo_target_dir):
|
46 |
+
print(f"Cloning TripoSG repository ({BRANCH} branch) into {repo_target_dir}...")
|
47 |
+
try:
|
48 |
+
subprocess.run(["git", "clone", "--branch", BRANCH, "--depth", "1", repo_url_to_clone, repo_target_dir], check=True)
|
49 |
+
code_source_path = repo_target_dir
|
50 |
+
print("Repository cloned successfully.")
|
51 |
+
except subprocess.CalledProcessError as e:
|
52 |
+
print(f"Error cloning repository: {e}")
|
53 |
+
print("Please ensure the URL is correct, the branch '{BRANCH}' exists, and you have access rights (or provide a GITHUB_TOKEN).")
|
54 |
+
sys.exit(1)
|
55 |
+
except Exception as e:
|
56 |
+
print(f"An unexpected error occurred during cloning: {e}")
|
57 |
+
sys.exit(1)
|
58 |
+
else:
|
59 |
+
print(f"Directory {repo_target_dir} already exists. Assuming it contains the correct code/branch.")
|
60 |
+
# Optional: Add checks here like git pull or verifying the branch
|
61 |
+
code_source_path = repo_target_dir
|
62 |
+
|
63 |
+
if not code_source_path:
|
64 |
+
print("Error: Could not determine TripoSG code source path.")
|
65 |
+
sys.exit(1)
|
66 |
+
|
67 |
+
# Add repo to Python path
|
68 |
+
sys.path.insert(0, code_source_path) # Use the determined absolute path
|
69 |
+
print(f"Added {code_source_path} to sys.path")
|
70 |
+
# --- End Repo Setup ---
|
71 |
+
|
72 |
+
# --- ZeroGPU Setup ---
|
73 |
+
ENABLE_ZEROGPU = os.environ.get("ENABLE_ZEROGPU", "false").lower() in ("true", "1", "t")
|
74 |
+
print(f"ZeroGPU Enabled: {ENABLE_ZEROGPU}")
|
75 |
+
# --- End ZeroGPU Setup ---
|
76 |
+
|
77 |
+
if ENABLE_ZEROGPU:
|
78 |
+
import spaces # Import spaces for ZeroGPU
|
79 |
+
from PIL import Image
|
80 |
+
import numpy as np
|
81 |
+
import torch
|
82 |
+
from triposg.pipelines.pipeline_triposg_scribble import TripoSGScribblePipeline
|
83 |
+
import tempfile
|
84 |
+
|
85 |
+
# --- Weight Loading Logic ---
|
86 |
+
HF_TOKEN = os.environ.get("HF_TOKEN")
|
87 |
+
if HF_TOKEN:
|
88 |
+
HfFolder.save_token(HF_TOKEN)
|
89 |
+
HUGGING_FACE_REPO_ID = "VAST-AI/TripoSG-scribble"
|
90 |
+
DEFAULT_CACHE_PATH = "./pretrained_weights/TripoSG-scribble"
|
91 |
+
|
92 |
+
# Option 1: Use local path if WEIGHTS_PATH env var is set
|
93 |
+
local_weights_path = os.environ.get("WEIGHTS_PATH")
|
94 |
+
model_load_path = None
|
95 |
+
|
96 |
+
if local_weights_path:
|
97 |
+
print(f"Attempting to load weights from local path specified by WEIGHTS_PATH: {local_weights_path}")
|
98 |
+
if os.path.isdir(local_weights_path):
|
99 |
+
model_load_path = local_weights_path
|
100 |
+
print(f"Using local weights directory: {model_load_path}")
|
101 |
+
else:
|
102 |
+
print(f"Warning: WEIGHTS_PATH '{local_weights_path}' not found or not a directory. Falling back to Hugging Face download.")
|
103 |
+
|
104 |
+
# Option 2: Download from Hugging Face (if local path not used or invalid)
|
105 |
+
if not model_load_path:
|
106 |
+
hf_token = os.environ.get("HF_TOKEN")
|
107 |
+
print(f"Attempting to download weights from Hugging Face repo: {HUGGING_FACE_REPO_ID}")
|
108 |
+
if hf_token:
|
109 |
+
print("Using Hugging Face token for download.")
|
110 |
+
auth_token = hf_token
|
111 |
+
else:
|
112 |
+
print("No Hugging Face token found. Attempting public download.")
|
113 |
+
auth_token = None
|
114 |
+
try:
|
115 |
+
model_load_path = snapshot_download(
|
116 |
+
repo_id=HUGGING_FACE_REPO_ID,
|
117 |
+
local_dir=DEFAULT_CACHE_PATH,
|
118 |
+
local_dir_use_symlinks=False, # Recommended for Spaces
|
119 |
+
token=auth_token,
|
120 |
+
# revision="main" # Specify branch/commit if needed
|
121 |
+
)
|
122 |
+
print(f"Weights downloaded/cached to: {model_load_path}")
|
123 |
+
except Exception as e:
|
124 |
+
print(f"Error downloading weights from Hugging Face: {e}")
|
125 |
+
print("Please ensure the repository exists and is accessible, or provide a valid WEIGHTS_PATH.")
|
126 |
+
sys.exit(1) # Exit if weights cannot be loaded
|
127 |
+
|
128 |
+
# Load the pipeline using the determined path
|
129 |
+
print(f"Loading pipeline from: {model_load_path}")
|
130 |
+
pipe = TripoSGScribblePipeline.from_pretrained(model_load_path)
|
131 |
+
pipe.to(dtype=torch.float16, device="cuda")
|
132 |
+
print("Pipeline loaded.")
|
133 |
+
# --- End Weight Loading Logic ---
|
134 |
+
|
135 |
+
# Create a white background image and a transparent layer for drawing
|
136 |
+
canvas_width, canvas_height = 512, 512
|
137 |
+
initial_background = Image.new("RGB", (canvas_width, canvas_height), color="white")
|
138 |
+
initial_layer = Image.new("RGBA", (canvas_width, canvas_height), color=(0, 0, 0, 0)) # Transparent layer
|
139 |
+
# Prepare the initial value dictionary for ImageEditor
|
140 |
+
initial_value = {
|
141 |
+
"background": initial_background,
|
142 |
+
"layers": [initial_layer], # Add the transparent layer
|
143 |
+
"composite": None
|
144 |
+
}
|
145 |
+
|
146 |
+
# --- ZeroGPU Setup ---
|
147 |
+
# ... existing ZeroGPU setup ...
|
148 |
+
|
149 |
+
MAX_SEED = np.iinfo(np.int32).max
|
150 |
+
|
151 |
+
def get_random_seed():
|
152 |
+
return random.randint(0, MAX_SEED)
|
153 |
+
|
154 |
+
# Apply decorator conditionally
|
155 |
+
@spaces.GPU(duration=120) if ENABLE_ZEROGPU else lambda func: func
|
156 |
+
def generate_3d(scribble_image_dict, prompt, scribble_confidence, seed): # Added seed parameter back
|
157 |
+
print("Generating 3D model...")
|
158 |
+
# Extract the composite image from the ImageEditor dictionary
|
159 |
+
if scribble_image_dict is None or scribble_image_dict.get("composite") is None:
|
160 |
+
print("No scribble image provided.")
|
161 |
+
return None # Return None if no image is provided
|
162 |
+
|
163 |
+
# --- Seed Handling ---
|
164 |
+
current_seed = int(seed)
|
165 |
+
print(f"Using seed: {current_seed}")
|
166 |
+
# --- End Seed Handling ---
|
167 |
+
|
168 |
+
# Get the composite image which includes the drawing
|
169 |
+
# The composite might be RGBA if a layer was involved, ensure RGB for processing
|
170 |
+
image = Image.fromarray(scribble_image_dict["composite"]).convert("RGB")
|
171 |
+
|
172 |
+
# Preprocess the image: invert colors (black on white -> white on black)
|
173 |
+
image_np = np.array(image)
|
174 |
+
processed_image_np = 255 - image_np
|
175 |
+
processed_image = Image.fromarray(processed_image_np)
|
176 |
+
print("Image preprocessed.")
|
177 |
+
|
178 |
+
# Define fixed parameters
|
179 |
+
attn_scale_text = 1.0 # As per the example run.py
|
180 |
+
|
181 |
+
# Set the generator with the provided seed
|
182 |
+
generator = torch.Generator(device='cuda').manual_seed(current_seed)
|
183 |
+
|
184 |
+
# Run the pipeline
|
185 |
+
print("Running pipeline...")
|
186 |
+
out = pipe(
|
187 |
+
processed_image,
|
188 |
+
prompt=prompt,
|
189 |
+
num_tokens=512,
|
190 |
+
guidance_scale=0,
|
191 |
+
num_inference_steps=16,
|
192 |
+
attention_kwargs={
|
193 |
+
"cross_attention_scale": attn_scale_text,
|
194 |
+
"cross_attention_2_scale": scribble_confidence
|
195 |
+
},
|
196 |
+
generator=generator,
|
197 |
+
use_flash_decoder=False,
|
198 |
+
dense_octree_depth=8,
|
199 |
+
hierarchical_octree_depth=8
|
200 |
+
)
|
201 |
+
print("Pipeline finished.")
|
202 |
+
|
203 |
+
# Save the output mesh to a temporary file
|
204 |
+
if out.meshes and len(out.meshes) > 0:
|
205 |
+
# Create a temporary file with .glb extension
|
206 |
+
with tempfile.NamedTemporaryFile(suffix=".glb", delete=False) as tmpfile:
|
207 |
+
output_path = tmpfile.name
|
208 |
+
out.meshes[0].export(output_path)
|
209 |
+
print(f"Mesh saved to temporary file: {output_path}")
|
210 |
+
return output_path
|
211 |
+
else:
|
212 |
+
print("Pipeline did not generate any meshes.")
|
213 |
+
return None
|
214 |
+
|
215 |
+
# Create the Gradio interface
|
216 |
+
with gr.Blocks() as demo:
|
217 |
+
gr.Markdown("# Scribble + Text to 3D Model Generator (TripoSG)")
|
218 |
+
gr.Markdown("Draw a scribble (black on white canvas), enter a text prompt, adjust confidence, set a seed, and generate a 3D model.") # Updated guidance
|
219 |
+
with gr.Row():
|
220 |
+
with gr.Column(scale=1):
|
221 |
+
image_input = gr.ImageEditor(
|
222 |
+
label="Scribble Input (Draw Black on White)",
|
223 |
+
value=initial_value,
|
224 |
+
image_mode="RGB",
|
225 |
+
brush=gr.Brush(default_color="#000000", color_mode="fixed", default_size=5), # Fixed small brush size
|
226 |
+
interactive=True,
|
227 |
+
eraser=gr.Brush(default_color="#FFFFFF", color_mode="fixed", default_size=20) # Fixed small eraser size
|
228 |
+
)
|
229 |
+
prompt_input = gr.Textbox(label="Prompt", placeholder="e.g., a cute cat wearing a hat")
|
230 |
+
confidence_input = gr.Slider(minimum=0.0, maximum=1.0, value=0.4, step=0.05, label="Scribble Confidence (attn_scale_image)")
|
231 |
+
seed_input = gr.Number(label="Seed", value=0, precision=0) # Added Seed input back
|
232 |
+
with gr.Row():
|
233 |
+
submit_button = gr.Button("Generate 3D Model", variant="primary", scale=1)
|
234 |
+
lucky_button = gr.Button("I'm Feeling Lucky", scale=1)
|
235 |
+
with gr.Column(scale=1):
|
236 |
+
model_output = gr.Model3D(label="Generated 3D Model", interactive=False)
|
237 |
+
|
238 |
+
# Define the inputs for the main generation function
|
239 |
+
gen_inputs = [image_input, prompt_input, confidence_input, seed_input]
|
240 |
+
|
241 |
+
submit_button.click(
|
242 |
+
fn=generate_3d,
|
243 |
+
inputs=gen_inputs, # Include seed_input
|
244 |
+
outputs=model_output
|
245 |
+
)
|
246 |
+
|
247 |
+
# Define inputs for the lucky button (same as main button for the final call)
|
248 |
+
lucky_gen_inputs = [image_input, prompt_input, confidence_input, seed_input]
|
249 |
+
|
250 |
+
lucky_button.click(
|
251 |
+
fn=get_random_seed, # First, get a random seed
|
252 |
+
inputs=[],
|
253 |
+
outputs=[seed_input] # Update the seed input field
|
254 |
+
).then(
|
255 |
+
fn=generate_3d, # Then, generate the model
|
256 |
+
inputs=lucky_gen_inputs, # Use the updated seed from the input field
|
257 |
+
outputs=model_output
|
258 |
+
)
|
259 |
+
|
260 |
+
# Launch with queue enabled if using ZeroGPU
|
261 |
+
print("Launching Gradio interface...")
|
262 |
+
demo.launch(share=False, server_name="0.0.0.0")
|
263 |
+
print("Gradio interface launched.")
|
requirements.txt
ADDED
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
diffusers
|
2 |
+
transformers==4.49.0
|
3 |
+
einops
|
4 |
+
huggingface_hub
|
5 |
+
opencv-python
|
6 |
+
trimesh==4.5.3
|
7 |
+
omegaconf
|
8 |
+
scikit-image
|
9 |
+
numpy
|
10 |
+
peft
|
11 |
+
scipy==1.11.4
|
12 |
+
jaxtyping
|
13 |
+
typeguard
|
14 |
+
ninja
|
15 |
+
gltflib
|
16 |
+
https://huggingface.co/spaces/VAST-AI/TripoSG/resolve/main/diso-0.1.4-cp310-cp310-linux_x86_64.whl?download=true
|