Inference API Reference

Model inference and visualization.

Module: inference.predictor

Class: AutoStructN2VPredictor

Predictor for applying trained models to denoise images.

Constructor:

AutoStructN2VPredictor(model, device=None, patch_size=64, stride=None)

Parameters:

• model (nn.Module): Trained model
• device (torch.device, optional): Inference device (auto-detect if None)
• patch_size (int): Patch size (default: 64)
• stride (int, optional): Stride for patches (default: patch_size//2)

Methods:

denoise_image(image_path, output_path=None, show=False)

Denoise single image.

Parameters:

• image_path (str): Input image path
• output_path (str, optional): Output path
• show (bool): Display result (default: False)

Returns:

• numpy.ndarray: Denoised image

Example:

from autoStructN2V.inference import AutoStructN2VPredictor

predictor = AutoStructN2VPredictor(model=model, patch_size=64)
denoised = predictor.denoise_image('noisy.tif', 'denoised.tif', show=True)

denoise_tensor(img_tensor)

Denoise image tensor using patch-based processing.

Parameters:

• img_tensor (torch.Tensor): Input tensor (B, C, H, W)

Returns:

• torch.Tensor: Denoised tensor

process_directory(input_dir, output_dir=None, show=False)

Process all TIFF images in directory.

Parameters:

• input_dir (str): Input directory
• output_dir (str, optional): Output directory
• show (bool): Display results (default: False)

Returns:

• list: Output image paths

Example:

output_paths = predictor.process_directory(
    input_dir='./noisy_images/',
    output_dir='./denoised_outputs/'
)
print(f"Processed {len(output_paths)} images")

from_checkpoint(checkpoint_path, model_class, stage, **kwargs) [Class Method]

Create predictor from checkpoint.

Parameters:

• checkpoint_path (str): Checkpoint path
• model_class: Model class
• stage (str): 'stage1' or 'stage2'
• **kwargs: Additional predictor args

Returns:

• AutoStructN2VPredictor

Example:

predictor = AutoStructN2VPredictor.from_checkpoint(
    checkpoint_path='stage2_model.pth',
    model_class=AutoStructN2VModel,
    stage='stage2',
    patch_size=64
)

---

Module: inference.visualization

Function: visualize_denoising_result

Visualize original vs denoised with histograms.

Signature:

visualize_denoising_result(original, denoised, title=None, 
                           figsize=(12, 6), cmap='gray')

Parameters:

• original (np.ndarray): Original image
• denoised (np.ndarray): Denoised image
• title (str, optional): Figure title
• figsize (tuple): Figure size
• cmap (str): Colormap (default: 'gray')

Returns:

• matplotlib.figure.Figure

Example:

from autoStructN2V.inference import visualize_denoising_result
from autoStructN2V.utils.image import load_and_normalize_image

original = load_and_normalize_image('original.tif')
denoised = load_and_normalize_image('denoised.tif')

fig = visualize_denoising_result(original, denoised, title="Results")
fig.savefig('comparison.png')

---

Function: compare_multiple_images

Compare multiple images in grid layout.

Signature:

compare_multiple_images(image_arrays, titles=None, 
                        figsize=(15, 10), cmap='gray')

Parameters:

• image_arrays (list): List of image arrays
• titles (list, optional): Image titles
• figsize (tuple): Figure size
• cmap (str): Colormap

Returns:

• matplotlib.figure.Figure

Example:

from autoStructN2V.inference import compare_multiple_images

images = [original, stage1_output, stage2_output]
titles = ['Original', 'Stage 1', 'Stage 2']

fig = compare_multiple_images(images, titles)

---

Function: create_difference_map

Create difference map between images.

Signature:

create_difference_map(original, denoised, enhanced=True)

Parameters:

• original (np.ndarray): Original image
• denoised (np.ndarray): Denoised image
• enhanced (bool): Enhance contrast (default: True)

Returns:

• numpy.ndarray: Difference map

---

See Also:

• Inference Guide
• Basic Usage Tutorial