mmcv和openCV两个库imcrop()和imresize()方法的对应【基础分析】

news2024/11/23 2:56:32

🥇 版权: 本文由【墨理学AI】原创首发、各位读者大大、敬请查阅、感谢三连
🎉 声明: 作为全网 AI 领域干货最多的博主之一，❤️ 不负光阴不负卿 ❤️

0-9

文章目录

- MMCV 全家桶
- mmcv.imresize(img, (1000, 600), return_scale=True) 方法实现
- 对应的 openCV 的 cv2.resize 方法
- mmcv 中 imcrop(img, bboxes, scale=1.0, pad_fill=None) 方法
- mmcv/image/geometric.py 文件分析
- 📙 精选专栏

MMCV 全家桶

https://github.com/open-mmlab/mmcv/blob/master/README_zh-CN.md

是一个用于训练深度学习模型的基础库

1-90

有很多star数量较高的集成仓库方便开发者进行模型训练和部署的全流程搭建

1-9

mmcv.imresize(img, (1000, 600), return_scale=True) 方法实现

mmcv.imresize(img, (1000, 600), return_scale=True)

源码链接如下

https://github.com/open-mmlab/mmcv/blob/b9bdfeb6f6aa67372d18b2ed5a10de28fc179d11/mmcv/image/geometric.py


cv2_interp_codes = {
    'nearest': cv2.INTER_NEAREST,
    'bilinear': cv2.INTER_LINEAR,
    'bicubic': cv2.INTER_CUBIC,
    'area': cv2.INTER_AREA,
    'lanczos': cv2.INTER_LANCZOS4
}

def imresize(img,
             size,
             return_scale=False,
             interpolation='bilinear',
             out=None,
             backend=None):
    """Resize image to a given size.
    Args:
        img (ndarray): The input image.
        size (tuple[int]): Target size (w, h).
        return_scale (bool): Whether to return `w_scale` and `h_scale`.
        interpolation (str): Interpolation method, accepted values are
            "nearest", "bilinear", "bicubic", "area", "lanczos" for 'cv2'
            backend, "nearest", "bilinear" for 'pillow' backend.
        out (ndarray): The output destination.
        backend (str | None): The image resize backend type. Options are `cv2`,
            `pillow`, `None`. If backend is None, the global imread_backend
            specified by ``mmcv.use_backend()`` will be used. Default: None.
    Returns:
        tuple | ndarray: (`resized_img`, `w_scale`, `h_scale`) or
        `resized_img`.
    """
    h, w = img.shape[:2]
    if backend is None:
        backend = imread_backend
    if backend not in ['cv2', 'pillow']:
        raise ValueError(f'backend: {backend} is not supported for resize.'
                         f"Supported backends are 'cv2', 'pillow'")

    if backend == 'pillow':
        assert img.dtype == np.uint8, 'Pillow backend only support uint8 type'
        pil_image = Image.fromarray(img)
        pil_image = pil_image.resize(size, pillow_interp_codes[interpolation])
        resized_img = np.array(pil_image)
    else:
        resized_img = cv2.resize(
            img, size, dst=out, interpolation=cv2_interp_codes[interpolation])
    if not return_scale:
        return resized_img
    else:
        w_scale = size[0] / w
        h_scale = size[1] / h
        return resized_img, w_scale,

对应的 openCV 的 cv2.resize 方法

resized_img = cv2.resize(
            img, size, dst=out, interpolation=cv2_interp_codes[interpolation])

mmcv 中 imcrop(img, bboxes, scale=1.0, pad_fill=None) 方法

调用方式

bboxes = np.array([10, 10, 100, 120])
patch = mmcv.imcrop(img, bboxes)

mmcv/image/geometric.py 源码实现

def imcrop(img, bboxes, scale=1.0, pad_fill=None):
    """Crop image patches.
    3 steps: scale the bboxes -> clip bboxes -> crop and pad.
    Args:
        img (ndarray): Image to be cropped.
        bboxes (ndarray): Shape (k, 4) or (4, ), location of cropped bboxes.
        scale (float, optional): Scale ratio of bboxes, the default value
            1.0 means no padding.
        pad_fill (Number | list[Number]): Value to be filled for padding.
            Default: None, which means no padding.
    Returns:
        list[ndarray] | ndarray: The cropped image patches.
    """
    chn = 1 if img.ndim == 2 else img.shape[2]
    if pad_fill is not None:
        if isinstance(pad_fill, (int, float)):
            pad_fill = [pad_fill for _ in range(chn)]
        assert len(pad_fill) == chn

    _bboxes = bboxes[None, ...] if bboxes.ndim == 1 else bboxes
    scaled_bboxes = bbox_scaling(_bboxes, scale).astype(np.int32)
    clipped_bbox = bbox_clip(scaled_bboxes, img.shape)

    patches = []
    for i in range(clipped_bbox.shape[0]):
        x1, y1, x2, y2 = tuple(clipped_bbox[i, :])
        if pad_fill is None:
            patch = img[y1:y2 + 1, x1:x2 + 1, ...]
        else:
            _x1, _y1, _x2, _y2 = tuple(scaled_bboxes[i, :])
            if chn == 1:
                patch_shape = (_y2 - _y1 + 1, _x2 - _x1 + 1)
            else:
                patch_shape = (_y2 - _y1 + 1, _x2 - _x1 + 1, chn)
            patch = np.array(
                pad_fill, dtype=img.dtype) * np.ones(
                    patch_shape, dtype=img.dtype)
            x_start = 0 if _x1 >= 0 else -_x1
            y_start = 0 if _y1 >= 0 else -_y1
            w = x2 - x1 + 1
            h = y2 - y1 + 1
            patch[y_start:y_start + h, x_start:x_start + w,
                  ...] = img[y1:y1 + h, x1:x1 + w, ...]
        patches.append(patch)

    if bboxes.ndim == 1:
        return patches[0]
    else:
        return patches

依赖方法

def bbox_clip(bboxes, img_shape):
    """Clip bboxes to fit the image shape.
    Args:
        bboxes (ndarray): Shape (..., 4*k)
        img_shape (tuple[int]): (height, width) of the image.
    Returns:
        ndarray: Clipped bboxes.
    """
    assert bboxes.shape[-1] % 4 == 0
    cmin = np.empty(bboxes.shape[-1], dtype=bboxes.dtype)
    cmin[0::2] = img_shape[1] - 1
    cmin[1::2] = img_shape[0] - 1
    clipped_bboxes = np.maximum(np.minimum(bboxes, cmin), 0)
    return clipped_bboxes

def bbox_scaling(bboxes, scale, clip_shape=None):
    """Scaling bboxes w.r.t the box center.
    Args:
        bboxes (ndarray): Shape(..., 4).
        scale (float): Scaling factor.
        clip_shape (tuple[int], optional): If specified, bboxes that exceed the
            boundary will be clipped according to the given shape (h, w).
    Returns:
        ndarray: Scaled bboxes.
    """
    if float(scale) == 1.0:
        scaled_bboxes = bboxes.copy()
    else:
        w = bboxes[..., 2] - bboxes[..., 0] + 1
        h = bboxes[..., 3] - bboxes[..., 1] + 1
        dw = (w * (scale - 1)) * 0.5
        dh = (h * (scale - 1)) * 0.5
        scaled_bboxes = bboxes + np.stack((-dw, -dh, dw, dh), axis=-1)
    if clip_shape is not None:
        return bbox_clip(scaled_bboxes, clip_shape)
    else:
        return scaled_bboxes

mmcv/image/geometric.py 文件分析

从头文件的import和基础代码分析，可以看到 mmcv 中是把 openCV 和 PIL 两个库的 image 处理集成到 geometric.py 各个方法中

import numbers
import warnings
from typing import Optional, Tuple

import cv2
import numpy as np

from ..utils import to_2tuple
from .io import imread_backend

try:
    from PIL import Image
except ImportError:
    Image = None