aniket47/text-to-3d-backend

Merge branch 'main' of https://huggingface.co/spaces/aniket47/text-to-3d-backend

Restore backend repository with latest fixes

Fix trimesh index out of bounds error: Resolve dimension mismatch between image and depth map - Resize image to match depth map dimensions exactly - Add bounds checking for array access - Improve face generation with validation - Add detailed dimension logging for debugging

Critical fix: Add null path validation and complete trimesh fallback implementation - Fix 'stat: path should be NoneType' error by validating depth_result success - Add proper null checks for obj_path and depth_map_path - Implement missing _create_with_trimesh method - Improve error handling with detailed logging

Fix Cloudinary file upload issues: Add file existence checks, better error handling, and detailed logging for upload debugging

Remove Open3D dependencies and simplify for HF Spaces deployment - Use trimesh-only mode with FORCE_TRIMESH flag - Remove OpenGL system dependencies - Simplify requirements.txt and Dockerfile

Fix Open3D OpenGL dependencies for HuggingFace Spaces headless environment

MAJOR UPGRADE: Implement Open3D Poisson reconstruction from python-api for superior 3D quality

Major 3D quality improvement: implement proper camera projection and Delaunay triangulation from original code

Fix 3D model stretching: reduce depth scaling and improve proportions

Improve model reliability: add more fallback models, retry logic, and better error handling

Fix critical data format mismatch: add required success flag and image_pil field

Add enhanced debugging and error handling for image processing

Fix InferenceClient initialization: change api_key to token parameter

Add test script and commit remaining changes from InferenceClient migration

Add huggingface_hub dependency for InferenceClient migration

Fix Hugging Face Inference API implementation

Add .gitignore file for backend environment variables and cache files

Fix syntax errors in image_generator.py

Switch to API-based image generation with local depth processing

Implement robust CPU fallback mechanism for GPU errors

Fix CPU-only compatibility for HF Spaces free tier

Fix requirements.txt corruption: Restore proper package formatting

Fix PyTorch dependency conflict: Update to version 2.1.2

Fix build error: Remove build-time model preloading and optimize for HF Spaces

Fix dependency conflict: Update xformers to 0.0.23.post1 for torch 2.1.1 compatibility

Speed optimizations: pre-download models, optimized caching, model CPU offloading

Upgrade to Stability AI SDXL model for superior image quality

Restore high-quality image generation settings - 512x512 resolution, 25 inference steps

Add fallback image generation for timeout scenarios

Optimize image generation for CPU - reduce steps, add timeout, smaller images for CPU

Add results endpoint for proper job completion handling

Final update to use Intel DPT BEIT Large 512 model as requested

Use Intel DPT BEIT Large 512 model as requested

Fix CPU-only compatibility - remove GPU offloading on CPU-only systems

Complete migration from open3d to trimesh for HF Spaces compatibility

Restore full FastAPI with ML models

Add Dockerfile for HF Spaces Docker deployment

Switch to minimal FastAPI for testing deployment

Initial FastAPI backend for HF Spaces