digiPal / models /model_3d_generator.py
BladeSzaSza's picture
try to install everything at first
5af079f
import torch
import numpy as np
from PIL import Image
import trimesh
import tempfile
from typing import Union, Optional, Dict, Any
from pathlib import Path
import os
import logging
import random
import time
import threading
from huggingface_hub import snapshot_download
import shutil
# Set up detailed logging for 3D generation
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class TimeoutError(Exception):
"""Custom timeout exception"""
pass
class Hunyuan3DGenerator:
"""3D model generation using Hunyuan3D-2.1 directly"""
def __init__(self, device: str = "cuda"):
logger.info(f"πŸ”§ Initializing Hunyuan3DGenerator with device: {device}")
self.device = device if torch.cuda.is_available() else "cpu"
logger.info(f"πŸ”§ Final device selection: {self.device}")
self.model = None
self.preprocessor = None
# Model configuration
self.model_id = "tencent/Hunyuan3D-2.1"
self.model_path = None
# Generation parameters
self.num_inference_steps = 30 # Reduced for faster generation
self.guidance_scale = 7.5
self.resolution = 256 # 3D resolution
# Timeout configuration
self.generation_timeout = 180 # 3 minutes timeout for local generation
# Use full model since we have enough RAM
logger.info(f"πŸ”§ Using full Hunyuan3D-2.1 model")
logger.info(f"⏱️ Generation timeout set to: {self.generation_timeout} seconds")
def _check_vram(self) -> bool:
"""Check if we have enough VRAM for full model"""
logger.info("πŸ” Checking VRAM availability...")
if not torch.cuda.is_available():
logger.info("❌ CUDA not available")
return False
try:
vram = torch.cuda.get_device_properties(0).total_memory
vram_gb = vram / (1024 * 1024 * 1024)
logger.info(f"πŸ” Available VRAM: {vram_gb:.2f} GB")
# Need at least 12GB for full model
has_enough = vram > 12 * 1024 * 1024 * 1024
logger.info(f"πŸ” Has enough VRAM (>12GB): {has_enough}")
return has_enough
except Exception as e:
logger.error(f"❌ Error checking VRAM: {e}")
return False
def load_model(self):
"""Load Hunyuan3D model and run necessary setup"""
if self.model is None:
logger.info("πŸš€ Starting Hunyuan3D model loading and setup...")
try:
import subprocess
import sys
import os
def run_setup_command(command, cwd):
logger.info(f"Running command: {' '.join(command)} in {cwd}")
try:
process = subprocess.run(
command,
check=True,
capture_output=True,
text=True,
cwd=cwd
)
logger.info(f"βœ… Command successful.")
if process.stdout:
logger.info(f"STDOUT:\n{process.stdout}")
if process.stderr:
logger.warning(f"STDERR:\n{process.stderr}")
except subprocess.CalledProcessError as e:
logger.error(f"❌ Command failed with exit code {e.returncode}")
logger.error(f"STDOUT:\n{e.stdout}")
logger.error(f"STDERR:\n{e.stderr}")
raise # Re-raise the exception to halt execution and see the error
# Download model repository if not already present
logger.info(f"πŸ“₯ Downloading Hunyuan3D repository from {self.model_id}...")
self.model_path = snapshot_download(
repo_id=self.model_id,
repo_type="space",
cache_dir="./models/hunyuan3d_cache"
)
logger.info(f"βœ… Model repository downloaded to: {self.model_path}")
# # List the contents of the downloaded directory for debugging
# logger.info(f"πŸ” Listing contents of {self.model_path}...")
# run_setup_command(['ls', '-R'], cwd=self.model_path)
# --- Installation and Compilation ---
logger.info("πŸ”§ Running Hunyuan3D setup scripts with detailed logging...")
# 1. Install requirements from the model's specific requirements file
# requirements_path = os.path.join(self.model_path, 'requirements_hunyuan3d.txt')
# if os.path.exists(requirements_path):
# pip_command = [
# sys.executable, '-m', 'pip', 'install', '-r', requirements_path,
# '--extra-index-url', 'https://mirrors.cloud.tencent.com/pypi/simple/',
# '--extra-index-url', 'https://mirrors.aliyun.com/pypi/simple'
# ]
# run_setup_command(pip_command, cwd=self.model_path)
# 2. Install custom rasterizer dependencies (torch)
# logger.info("Installing torch, torchvision, torchaudio...")
# pip_command_torch = [sys.executable, '-m', 'pip', 'install', 'torch==2.5.1', 'torchvision==0.20.1', 'torchaudio==2.5.1', '--index-url', 'https://download.pytorch.org/whl/cu124']
# run_setup_command(pip_command_torch, cwd=self.model_path)
# 3. Install custom rasterizer
rasterizer_path = os.path.join(self.model_path, 'hy3dpaint', 'packages', 'custom_rasterizer')
if os.path.exists(rasterizer_path):
pip_command_rasterizer = [sys.executable, '-m', 'pip', 'install', '--no-build-isolation', '-e', '.']
run_setup_command(pip_command_rasterizer, cwd=rasterizer_path)
# 4. Compile mesh painter
renderer_path = os.path.join(self.model_path, 'hy3dpaint', 'DifferentiableRenderer')
compile_script_path = os.path.join(renderer_path, 'compile_mesh_painter.sh')
if os.path.exists(compile_script_path):
bash_command = ['bash', compile_script_path]
run_setup_command(bash_command, cwd=renderer_path)
logger.info("βœ… Hunyuan3D setup completed successfully.")
# --- Pipeline Initialization ---
logger.info("✈️ Initializing Hunyuan3D pipelines...")
# Add subdirectories to Python path
sys.path.insert(0, os.path.join(self.model_path, 'hy3dshape'))
sys.path.insert(0, os.path.join(self.model_path, 'hy3dpaint'))
# Import the correct pipelines
from hy3dshape.pipelines import Hunyuan3DDiTFlowMatchingPipeline
from textureGenPipeline import Hunyuan3DPaintPipeline, Hunyuan3DPaintConfig
# Instantiate pipelines
logger.info("Instantiating shape pipeline...")
self.shape_pipeline = Hunyuan3DDiTFlowMatchingPipeline.from_pretrained(
self.model_path, torch_dtype=torch.bfloat16
).to(self.device)
logger.info("Instantiating paint pipeline...")
paint_config = Hunyuan3DPaintConfig(max_num_view=8, resolution=1024, pbr_optimization=True)
self.paint_pipeline = Hunyuan3DPaintPipeline(paint_config)
self.model = "direct_model"
logger.info("βœ… Hunyuan3D pipelines loaded successfully.")
except Exception as e:
logger.error(f"❌ Failed to set up Hunyuan3D pipeline: {e}", exc_info=True)
logger.warning("πŸ”„ Falling back to simplified 3D generation...")
self.model = "simplified"
def image_to_3d(self,
image: Union[str, Image.Image, np.ndarray],
remove_background: bool = True,
texture_resolution: int = 1024) -> Union[str, trimesh.Trimesh]:
"""Convert 2D image to 3D model using local Hunyuan3D"""
logger.info("🎯 Starting image-to-3D conversion process...")
logger.info(f"🎯 Input type: {type(image)}")
logger.info(f"🎯 Remove background: {remove_background}")
logger.info(f"🎯 Texture resolution: {texture_resolution}")
try:
# Load model if needed
logger.info("πŸ” Checking if model needs loading...")
if self.model is None:
logger.info("πŸ“¦ Model not loaded, initiating loading...")
self.load_model()
else:
logger.info("βœ… Model already loaded")
# Prepare image
logger.info("πŸ–ΌοΈ Preparing input image...")
if isinstance(image, str):
logger.info(f"πŸ–ΌοΈ Loading image from path: {image}")
image = Image.open(image)
elif isinstance(image, np.ndarray):
logger.info("πŸ–ΌοΈ Converting numpy array to PIL Image")
image = Image.fromarray(image)
# Ensure image is PIL Image
if not isinstance(image, Image.Image):
logger.error("❌ Invalid image type")
raise ValueError("Image must be PIL Image, numpy array, or path string")
logger.info(f"πŸ–ΌοΈ Image mode: {image.mode}, size: {image.size}")
# Process based on model type
if self.model == "direct_model":
logger.info("🌐 Using direct Hunyuan3D model for 3D generation...")
return self._generate_with_direct_model(image, remove_background, texture_resolution)
elif self.model == "simplified":
logger.info("πŸ”„ Using simplified Hunyuan3D generation...")
return self._generate_simplified_3d(image)
else:
# Fallback to simple 3D generation
logger.info("πŸ”„ Using fallback 3D generation...")
return self._generate_fallback_3d(image)
except Exception as e:
logger.error(f"❌ 3D generation error: {e}")
logger.error(f"❌ Error type: {type(e).__name__}")
logger.info("πŸ”„ Falling back to simple 3D generation...")
return self._generate_fallback_3d(image)
def _generate_with_direct_model(self, image: Image.Image, remove_background: bool, texture_resolution: int) -> str:
"""Generate 3D model using the official Hunyuan3D pipelines"""
try:
# Remove background if requested
if remove_background:
logger.info("🎭 Removing background...")
image = self._remove_background(image)
# Save image to a temporary file, as pipelines expect a path
temp_image_path = self._save_temp_image(image)
# 1. Generate the untextured mesh
logger.info("πŸ”² Generating 3D shape with Hunyuan3DDiTFlowMatchingPipeline...")
# The pipeline returns a list of meshes, we take the first one
mesh_untextured_path = self.shape_pipeline(
image=temp_image_path,
num_inference_steps=self.num_inference_steps,
guidance_scale=self.guidance_scale,
seed=random.randint(1, 10000)
)[0]
logger.info(f"βœ… Untextured mesh saved to: {mesh_untextured_path}")
# 2. Generate the texture for the mesh
logger.info("🎨 Generating texture with Hunyuan3DPaintPipeline...")
mesh_textured_path = self.paint_pipeline(
mesh_path=mesh_untextured_path,
image_path=temp_image_path,
guidance_scale=self.guidance_scale,
seed=random.randint(1, 10000)
)
logger.info(f"βœ… Textured mesh saved to: {mesh_textured_path}")
# 3. Save the final output to a consistent location
output_path = self._save_output_mesh(mesh_textured_path)
logger.info(f"βœ… 3D model generation successful. Final model at: {output_path}")
return output_path
except Exception as e:
logger.error(f"❌ Direct model generation failed: {e}", exc_info=True)
raise
def _generate_simplified_3d(self, image: Image.Image) -> str:
"""Generate 3D using simplified approach with PyTorch operations"""
logger.info("πŸ”§ Using simplified 3D generation with PyTorch...")
try:
# Convert image to tensor
import torchvision.transforms as transforms
transform = transforms.Compose([
transforms.Resize((256, 256)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
])
image_tensor = transform(image).unsqueeze(0).to(self.device)
# Create a depth map from the image
logger.info("πŸ“ Generating depth map...")
# Simple depth estimation based on image brightness
gray_image = image.convert('L')
depth_array = np.array(gray_image.resize((64, 64))) / 255.0
# Apply some smoothing and scaling
from scipy.ndimage import gaussian_filter
depth_array = gaussian_filter(depth_array, sigma=2)
depth_array = depth_array * 0.3 + 0.1 # Scale depth
# Generate mesh from depth map
logger.info("πŸ”² Creating mesh from depth map...")
mesh = self._depthmap_to_mesh(depth_array, image)
# Save mesh
output_path = self._save_mesh(mesh)
logger.info(f"βœ… Simplified 3D model generated: {output_path}")
return output_path
except Exception as e:
logger.error(f"❌ Simplified generation failed: {e}")
return self._generate_fallback_3d(image)
def _depthmap_to_mesh(self, depth_map: np.ndarray, texture_image: Image.Image) -> trimesh.Trimesh:
"""Convert depth map to textured 3D mesh"""
h, w = depth_map.shape
# Create vertices with texture coordinates
vertices = []
faces = []
vertex_colors = []
# Resize texture to match depth map
texture_resized = texture_image.resize((w, h))
texture_array = np.array(texture_resized)
# Create vertex grid with colors
for i in range(h):
for j in range(w):
x = (j - w/2) / w * 2
y = (i - h/2) / h * 2
z = depth_map[i, j]
vertices.append([x, y, z])
# Add vertex color from texture
if len(texture_array.shape) == 3:
color = texture_array[i, j, :3]
else:
color = [texture_array[i, j]] * 3
vertex_colors.append(color)
# Create faces (two triangles per grid square)
for i in range(h-1):
for j in range(w-1):
v1 = i * w + j
v2 = v1 + 1
v3 = v1 + w
v4 = v3 + 1
faces.append([v1, v2, v3])
faces.append([v2, v4, v3])
vertices = np.array(vertices)
faces = np.array(faces)
vertex_colors = np.array(vertex_colors, dtype=np.uint8)
# Create mesh with vertex colors
mesh = trimesh.Trimesh(
vertices=vertices,
faces=faces,
vertex_colors=vertex_colors
)
# Apply smoothing
mesh = mesh.smoothed()
# Add a base to make it more stable
base_vertices, base_faces = self._create_base(vertices, w, h)
base_mesh = trimesh.Trimesh(vertices=base_vertices, faces=base_faces)
# Combine mesh with base
mesh = trimesh.util.concatenate([mesh, base_mesh])
return mesh
def _create_base(self, vertices: np.ndarray, w: int, h: int) -> tuple:
"""Create a base for the mesh"""
base_z = vertices[:, 2].min() - 0.1
base_vertices = []
base_faces = []
# Get boundary vertices - fix the indexing
boundary_indices = []
# Top edge (excluding corners)
for j in range(1, w-1):
boundary_indices.append(j)
# Right edge (including top-right corner)
for i in range(h):
boundary_indices.append(i * w + w - 1)
# Bottom edge (excluding bottom-right corner, going right to left)
for j in range(w-2, 0, -1):
boundary_indices.append((h-1) * w + j)
# Left edge (including bottom-left corner, going bottom to top)
for i in range(h-1, -1, -1):
boundary_indices.append(i * w)
# Remove duplicate indices (first and last should not be the same)
if boundary_indices and boundary_indices[0] == boundary_indices[-1]:
boundary_indices = boundary_indices[:-1]
# Create base vertices
start_idx = len(vertices)
for idx in boundary_indices:
if idx < len(vertices): # Safety check
v = vertices[idx].copy()
v[2] = base_z
base_vertices.append(v)
if not base_vertices:
# If no base vertices were created, return empty arrays
return np.array([]), np.array([])
# Create center vertex
center = np.mean(base_vertices, axis=0)
base_vertices.append(center)
center_idx = len(base_vertices) - 1
# Create base faces
for i in range(len(boundary_indices)):
next_i = (i + 1) % len(boundary_indices)
base_faces.append([
i,
next_i,
center_idx
])
return np.array(base_vertices), np.array(base_faces)
def _remove_background(self, image: Image.Image) -> Image.Image:
"""Remove background from image"""
try:
# Try using rembg if available
from rembg import remove
return remove(image)
except:
# Fallback: simple background removal
# Convert to RGBA
image = image.convert("RGBA")
# Simple white background removal
datas = image.getdata()
new_data = []
for item in datas:
# Remove white-ish backgrounds
if item[0] > 230 and item[1] > 230 and item[2] > 230:
new_data.append((255, 255, 255, 0))
else:
new_data.append(item)
image.putdata(new_data)
return image
def _generate_fallback_3d(self, image: Union[Image.Image, np.ndarray]) -> str:
"""Generate fallback 3D model when main model fails"""
# Create a simple 3D representation based on image
if isinstance(image, np.ndarray):
image = Image.fromarray(image)
elif isinstance(image, str):
image = Image.open(image)
# Analyze image for basic shape
image_array = np.array(image.resize((64, 64)))
# Create height map from image brightness
gray = np.mean(image_array, axis=2) if len(image_array.shape) == 3 else image_array
height_map = gray / 255.0
# Create mesh from height map
mesh = self._heightmap_to_mesh(height_map)
# Save and return path
return self._save_mesh(mesh)
def _heightmap_to_mesh(self, heightmap: np.ndarray) -> trimesh.Trimesh:
"""Convert heightmap to 3D mesh"""
h, w = heightmap.shape
# Create vertices
vertices = []
faces = []
# Create vertex grid
for i in range(h):
for j in range(w):
x = (j - w/2) / w * 2
y = (i - h/2) / h * 2
z = heightmap[i, j] * 0.5
vertices.append([x, y, z])
# Create faces
for i in range(h-1):
for j in range(w-1):
# Two triangles per grid square
v1 = i * w + j
v2 = v1 + 1
v3 = v1 + w
v4 = v3 + 1
faces.append([v1, v2, v3])
faces.append([v2, v4, v3])
vertices = np.array(vertices)
faces = np.array(faces)
# Create mesh
mesh = trimesh.Trimesh(vertices=vertices, faces=faces)
# Apply smoothing
mesh = mesh.smoothed()
return mesh
def _save_mesh(self, mesh: trimesh.Trimesh) -> str:
"""Save mesh to file"""
# Create temporary file
with tempfile.NamedTemporaryFile(suffix='.glb', delete=False) as tmp:
mesh_path = tmp.name
# Export mesh
mesh.export(mesh_path)
return mesh_path
def _save_temp_image(self, image: Image.Image) -> str:
"""Save PIL image to temporary file"""
with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:
image_path = tmp.name
# Save image
image.save(image_path, 'PNG')
logger.info(f"πŸ’Ύ Saved temp image to: {image_path}")
return image_path
def _save_output_mesh(self, source_mesh_path: str) -> str:
"""Copy generated mesh to our output location"""
# Create output directory if it doesn't exist
output_dir = "/tmp/hunyuan3d_output"
os.makedirs(output_dir, exist_ok=True)
# Generate unique filename
timestamp = tempfile.mktemp().split('/')[-1]
output_filename = f"hunyuan3d_mesh_{timestamp}.glb"
output_path = os.path.join(output_dir, output_filename)
# Copy the file
shutil.copy2(source_mesh_path, output_path)
logger.info(f"πŸ“ Copied mesh from {source_mesh_path} to {output_path}")
return output_path
def text_to_3d(self, text_prompt: str) -> str:
"""Generate 3D model from text description"""
# First generate image, then convert to 3D
# This would require image generator integration
raise NotImplementedError("Text to 3D requires image generation first")
def to(self, device: str):
"""Update device preference"""
self.device = device
logger.info(f"πŸ”§ Device preference updated to: {device}")
def __del__(self):
"""Cleanup when object is destroyed"""
if hasattr(self, 'model') and self.model not in [None, "fallback_mode", "simplified"]:
del self.model
if torch.cuda.is_available():
torch.cuda.empty_cache()