API: inference

Patch-based prediction and result visualization. Module: autoStructN2V.inference.

AutoStructN2VPredictor (inference/predictor.py)

AutoStructN2VPredictor(
    model, device=None,
    patch_size=64, stride=None,
    mode='2d',
    norm_stats=None,             # use the training run's stats for consistent scaling
    overlap_tile_pad=0,          # publication recipe: 16
)

Method	Purpose
denoise_stack(input_path, output_path=None, dtype='float32')	Multi-page TIFF in → denoised TIFF out
denoise_image(image)	Single 2D array
denoise_tensor(tensor)	Torch tensor in/out
process_directory(...)	Batch over a directory
AutoStructN2VPredictor.from_checkpoint(path, ...)	Rebuild model + predictor from a saved checkpoint

Slices are tiled into patches (with overlap_tile_pad context that is cropped from the output to avoid seams), predicted, and reassembled with weighted blending.

Visualization (inference/visualization.py)

• visualize_denoising_result(noisy, denoised, ...) — side-by-side panels
• compare_multiple_images(...) — grid comparison across methods/runs
• create_difference_map(noisy, denoised) — the residual; it should look like noise, with no biological structure

Example

from autoStructN2V.inference import AutoStructN2VPredictor

pred = AutoStructN2VPredictor.from_checkpoint(
    "results/my_run/model/model.pth", patch_size=64, overlap_tile_pad=16)
pred.denoise_stack("new_stack.tif", output_path="new_stack_denoised.tif")

See the Inference user guide for guidance on matching acquisition conditions.