pydiffbmp 0.1.0

DiffBMP: Fast Differentiable Painting with Any Image Primitives

DiffBMP

Overview

DiffBMP provides tools and methodologies for fast differentiable painting with any image primitives. It includes initialization techniques, rendering methods, and evaluation scripts aimed at comparing the effectiveness and quality of various vector art generation algorithms.

Installation

From PyPI (Recommended)

pip install pydiffbmp

System Requirements:

• Python: 3.8, 3.9, 3.10, or 3.11
• Platform: Linux x86_64 (pre-built CUDA binaries included)
• GPU: NVIDIA GPU with CUDA capability 8.6+ (RTX 3090, A100, RTX 4090, L40S, etc.)
• CUDA Driver: 11.8+ (no CUDA toolkit installation required)
• PyTorch: 1.13.0+

Additional system dependencies:

sudo apt-get install poppler-utils

From Source (Development)

For development or if you need to customize CUDA kernels:

git clone https://github.com/smhongok/diffbmp.git
cd diffbmp
pip install -r requirements.txt

# Build CUDA extensions (requires CUDA toolkit 12.3+)
./build_wheels.sh

# Install in editable mode
pip install -e .

Build Requirements (only for source installation):

• CUDA Toolkit 12.3+ (recommended)
• C++ compiler (GCC 9.4.0+ recommended)
• PyTorch 1.13.0+

Usage

Building

Tips :

• Recommended CUDA version is 12.3.
• Recommended C++ version is 9.4.0.

To build:

cd cuda_tile_rasterizer && python setup.py clean && rm -rf build/ *.egg-info *.so && python setup.py build_ext --inplace && cd ..

To build for fp16:

cd cuda_tile_rasterizer && python setup_fp16.py clean && rm -rf build/ *.egg-info *.so && python setup_fp16.py build_ext --inplace && cd ..

To build for psd_exporter:

cd cuda_tile_rasterizer && python setup_psd_export.py clean && rm -rf build/ *.egg-info *.so && python setup_psd_export.py build_ext --inplace && cd ..

If you want to specify the spec of your GPU or CUDA version:

cd cuda_tile_rasterizer && TORCH_CUDA_ARCH_LIST="8.6" CUDA_HOME=/usr/local/cuda-12.1 python setup_fp16.py clean && rm -rf build/ *.egg-info *.so && TORCH_CUDA_ARCH_LIST="8.6" CUDA_HOME=/usr/local/cuda-12.1 python setup_fp16.py build_ext --inplace && cd ..

If you have some errors when you build, remove followings and do above commands again: cuda_tile_rasterizer/cuda_tile_rasterizer, cuda_tile_rasterizer/cuda_tile_rasterizer_fp16, cuda_tile_rasterizer/build

Running the Main Script

Execute the main script with a configuration file:

python main.py --config configs/default.json

Evaluating Methods

To compare different circle-art methods:

python compare_methods.py --config configs/default.json

Running Evaluations

To execute specific evaluations on generated results:

python run_evaluation.py

Dynamic DiffBMP Visualization Methods

To visualize frame-by-frame primitive (x,y) movement:

paste the code below in configs/sequential.json

sequential_debug = {
    "route_visualization": {
        "enabled": True,
        "export_path": "./outputs/seq_test"
    }
}

To visualize per-pixel gradient of opacity-reduced primitives (by our heuristic):

paste the code below in configs/sequential.json

sequential_debug = {
    "gradient_visualization": {
        "enabled": True,
        "enable_non_problematic_primitive": False,
        "gradient_threshold": 1e-15,
        "save_dir": "./outputs/vis_class/debug_gradients_sequential"
    }
}

To visualize difference mask between frames (which becomes critical input for our heuristic):

paste the code below in configs/sequential.json

sequential_debug = {
    "diff_mask": {
        "enabled": True,
        "export_path": "./outputs/vis_class/diff_mask_sequential"
    }
}

Complete debugging config:

sequential_debug = {
    "gradient_visualization": {
        "enabled": True,
        "enable_non_problematic_primitive": False,
        "gradient_threshold": 1e-15,
        "save_dir": "./outputs/vis_class/debug_gradients_sequential"
    },
    "diff_mask": {
        "enabled": True,
        "export_path": "./outputs/vis_class/diff_mask_sequential"
    },
    "route_visualization": {
        "enabled": True,
        "export_path": "./outputs/seq_test"
    }
}

Assets

• Put any predefined SVG templates based on 'path' tag in the assets/svg directory.
• Put fonts for rendering in assets/font.

Examples and Datasets

The images directory contains various datasets and sample images categorized for quick testing:

• Artwork
• Nature
• Movie Posters
• Benchmark images (BSDS500, CelebA)

Contributing

Feel free to submit pull requests or report issues to enhance the functionality or resolve problems.

License

Please do not distribute. This is for the purpose of anonymous review.

---

Enjoy creating beautiful SVG drawing artwork!

Testing

To test all default*.json configs:

python test_configs.py --gpu 6 --no-wandb