Create exponential decay distribution.

Forward pass for single stroke.

Generate an image from text prompt if it doesn't already exist.

Create mapping from color names to normalized RGB values.

Get dictionary of data to save in checkpoint. Returns: Dict[str, Any]: Checkpoint data

Create a truncated normal distribution. This utility function creates a scipy truncated normal distribution with the specified parameter

Get timestep weight. Args: timestep: Current timestep Returns: float: Weight value

Handle raw image https://github.com/BachiLi/diffvg/blob/85802a71fbcc72d79cb75716eb4da4392fd09532/apps/refine_svg.py#L64

Initialize loss functions. Returns: Dict[str, nn.Module]: Dictionary of loss modules

Initialize the sketch model. Returns: SketchModel: Initialized model

Initialize the optimizer. Returns: torch.optim.Optimizer: Initialized optimizer Raises: ValueError: If optim

Initialize the learning rate scheduler. Returns: LRScheduler: Initialized scheduler Raises: ValueError: If s

Initialize Weights & Biases logging.

Check if current step is the final training step. Returns: bool: True if current step is final

Load image without segmentation. Args: image_path: Input image path transform: Transform pipeline Returns: tuple: (i

Log configuration parameters.

Save git repository information.

Save text prompt to file.

Main training function.

Normalize input to [-1, 1] range.

Plot batch with layer SVGs.

Plot batch with dropout outputs.

Run the pretraining phase of the model.

Sample diffusion timesteps. Args: train_step: Current training step batch_size: Number of samples Returns:

Sample truncation index.

Set random seeds for reproducibility.

Segment foreground object from image.

Plot single image.

Check if it's time to perform an interval-based action. Args: interval (int): Interval to check Returns: boo

Apply augmentations to input.

Calculate learning rate for current step. Args: step (int): Current step Returns: float: Learning rate

Initialize the training coach. Args: cfg: Configuration object model: Optional pre-initialized model

Initialize the learning rate scheduler. Args: func_type (str): Type of scheduling function max_steps (int): Maximum n

Initialize SDS loss. Args: cfg: Configuration object device: Device to run on

Initialize logger. Args: cfg: Configuration object

Initialize multi-output NeRF MLP.

Initialize Gaussian embedding.

Initialize RGB Gaussian embedding.

Initialize renderer with config and device.

Initialize the Painter model.

Initialize the base model.

Initialize and validate experiment directories.

Validate configuration after initialization.

Validate configuration after initialization.

Initialize and validate relationships between configurations.

Validate configuration after initialization.

Validate configuration after initialization.

Get color toggle values for current step.

Process a single color during training. Args: toggle_color_value: Color value to condition on Returns: tuple

Save visualization of current model state. Args: init_sketch: Initial sketch for comparison

Update model parameters and logging. Args: loss_dict: Dictionary of loss components

Aggregate loss dictionaries. Args: agg_loss_dict: List of loss dicts Returns: Aggregated loss dict

r""" Args: query, key, value: map a query and a set of key-value pairs to an output. See "Attention Is All You Nee

Compute SDS loss.

Forward pass through network.

Forward pass through embedding.

Forward pass through RGB embedding.

Forward pass interface.

Run inference with trained model. Args: save_filename: Base filename for outputs output_dir: Directory to save output

Run inference with a single RGB color. Args: save_filename (str): Base filename for saving outputs output_dir (str):

Initialize network weights.

Load a CLIP model Parameters ---------- name : str A model name listed by `clip.available_models()`, or the path to a model check

Save a model checkpoint.

Save and log an image.

Log training metrics.

Save an SVG file.