Week 9 Interactive Session: Model Implementation & Deployment

Course Roadmap Mapping

This week’s work in the broader GFM plan.

Week	Stage	Focus	You will build (geogfm)	Library tools	Outcome
9	Stage 3: Apply & Deploy	Deployment & Inference	inference/{tiling.py, sliding_window.py}	numpy, rasterio windows	Sliding-window inference on a small scene

Weekly goals

• Implement sliding-window/tiling inference utilities
• Stitch predictions; sanity-check outputs
• Outline API/UI deployment considerations

Session Outline (and Tangled Code)

• Concepts → Components mapping
  • Sliding-window and tiling utilities → inference/*.py

Package inits

# geogfm.inference

1) Sliding-window helpers

from __future__ import annotations
import numpy as np
from typing import Iterator, Tuple

Array = np.ndarray

def window_slices(height: int, width: int, patch_size: int, stride: int) -> Iterator[Tuple[slice, slice]]:
    for r in range(0, height - patch_size + 1, stride):
        for c in range(0, width - patch_size + 1, stride):
            yield slice(r, r + patch_size), slice(c, c + patch_size)

2) Tiling inference (naive)

from __future__ import annotations
import numpy as np
from typing import Callable
from geogfm.inference.sliding_window import window_slices

Array = np.ndarray

def apply_tiled(model_apply: Callable[[Array], Array], image: Array, patch_size: int, stride: int) -> Array:
    """Apply a function over tiles and stitch back naively (no overlaps blending)."""
    bands, height, width = image.shape
    output = np.zeros_like(image)
    for rs, cs in window_slices(height, width, patch_size, stride):
        pred = model_apply(image[:, rs, cs])  # (C, P, P)
        output[:, rs, cs] = pred
    return output