| """Tiling helpers shared by training and inference.""" |
| from __future__ import annotations |
|
|
| import numpy as np |
|
|
| from .config import STRIDE, TILE |
|
|
|
|
| def tile_origins(h: int, w: int, tile: int = TILE, stride: int = STRIDE) -> list[tuple[int, int]]: |
| """Top-left corners of overlapping tiles covering the image.""" |
| def axis(n: int) -> list[int]: |
| if n <= tile: |
| return [0] |
| pos = list(range(0, n - tile + 1, stride)) |
| if pos[-1] != n - tile: |
| pos.append(n - tile) |
| return pos |
|
|
| return [(x, y) for y in axis(h) for x in axis(w)] |
|
|
|
|
| def pad_to(img: np.ndarray, tile: int = TILE) -> np.ndarray: |
| """Pad the bottom/right so an image smaller than one tile still works.""" |
| h, w = img.shape[:2] |
| ph, pw = max(0, tile - h), max(0, tile - w) |
| if ph == 0 and pw == 0: |
| return img |
| return np.pad(img, [(0, ph), (0, pw)] + [(0, 0)] * (img.ndim - 2), mode="edge") |
|
|
|
|
| def boxes_in_tile(boxes: np.ndarray, x0: int, y0: int, tile: int = TILE, |
| min_visible: float = 0.7) -> np.ndarray: |
| """Boxes clipped into tile-local coordinates, keeping mostly-visible ones.""" |
| if len(boxes) == 0: |
| return np.zeros((0, 4), np.float32) |
| b = boxes.astype(np.float32).copy() |
| area = np.maximum(1.0, (b[:, 2] - b[:, 0]) * (b[:, 3] - b[:, 1])) |
| b[:, [0, 2]] -= x0 |
| b[:, [1, 3]] -= y0 |
| c = b.copy() |
| c[:, [0, 2]] = np.clip(c[:, [0, 2]], 0, tile) |
| c[:, [1, 3]] = np.clip(c[:, [1, 3]], 0, tile) |
| vis = np.clip(c[:, 2] - c[:, 0], 0, None) * np.clip(c[:, 3] - c[:, 1], 0, None) |
| return c[vis / area >= min_visible] |
|
|
