model_watermark_detect/watermark_verify/process/yolox_pytorch_blackbox_process.py

"""
yolox基于pytorch框架的黑盒水印处理验证流程
"""
import os

import cv2
import numpy as np
import onnxruntime
from watermark_verify.process.general_process_define import BlackBoxWatermarkProcessDefine

from watermark_verify.tools import parse_qrcode_label_file
from watermark_verify.tools.evaluate_tool import calculate_ciou


class DetectionProcess(BlackBoxWatermarkProcessDefine):
    def __init__(self, model_filename):
        super(DetectionProcess, self).__init__(model_filename)

    def process(self) -> bool:
        """
        根据流程定义进行处理，并返回模型标签验证结果
        :return: 模型标签验证结果
        """
        # 获取权重文件，使用触发集进行模型推理, 将推理结果与触发集预先二维码保存位置进行比对，在误差范围内则进行下一步，否则返回False
        cls_image_mapping = parse_qrcode_label_file.parse_labels(self.qrcode_positions_file)
        accessed_cls = set()
        for cls, images in cls_image_mapping.items():
            for image in images:
                image_path = os.path.join(self.trigger_dir, image)
                detect_result = self.detect_secret_label(image_path, self.model_filename, self.qrcode_positions_file, (640, 640))
                if detect_result:
                    accessed_cls.add(cls)
                    break
        if not accessed_cls == set(cls_image_mapping.keys()):  # 所有的分类都检测出模型水印，模型水印检测结果为True
            return False

        verify_result = self.verify_label()  # 模型标签检测通过，进行标签验证
        return verify_result


    def preprocess_image(self, image_path, input_size, swap=(2, 0, 1)):
        """
        对输入图片进行预处理
        :param swap: 维度变换元组，默认按照(2，0，1)进行变换
        :param image_path: 图片路径
        :param input_size: 模型输入大小
        :return: 图片经过处理完成的ndarray
        """
        img = cv2.imread(image_path)
        if len(img.shape) == 3:
            padded_img = np.ones((input_size[0], input_size[1], 3), dtype=np.uint8) * 114
        else:
            padded_img = np.ones(input_size, dtype=np.uint8) * 114

        r = min(input_size[0] / img.shape[0], input_size[1] / img.shape[1])
        resized_img = cv2.resize(
            img,
            (int(img.shape[1] * r), int(img.shape[0] * r)),
            interpolation=cv2.INTER_LINEAR,
        ).astype(np.uint8)
        padded_img[: int(img.shape[0] * r), : int(img.shape[1] * r)] = resized_img

        padded_img = padded_img.transpose(swap).copy()
        padded_img = np.ascontiguousarray(padded_img, dtype=np.float32)
        height, width, channels = img.shape
        return padded_img, r, height, width, channels

    def detect_secret_label(self, image_path, model_file, watermark_txt, input_shape) -> bool:
        """
        对模型使用触发集进行检查，判断是否存在黑盒模型水印，如果对嵌入水印的图片样本正确率高于阈值，证明模型存在黑盒水印
        :param image_path: 输入图像路径
        :param model_file: 模型文件路径
        :param watermark_txt: 水印标签文件路径
        :param input_shape: 模型输入图像大小，tuple
        :return: 检测结果
        """
        img, ratio, height, width, channels = self.preprocess_image(image_path, input_shape)
        x_center, y_center, w, h, cls = parse_qrcode_label_file.load_watermark_info(watermark_txt, image_path)
        # 计算绝对坐标
        x1 = (x_center - w / 2) * width
        y1 = (y_center - h / 2) * height
        x2 = (x_center + w / 2) * width
        y2 = (y_center + h / 2) * height
        watermark_box = [x1, y1, x2, y2, cls]
        if len(watermark_box) == 0:
            return False

        session = onnxruntime.InferenceSession(model_file)

        ort_inputs = {session.get_inputs()[0].name: img[None, :, :, :]}
        output = session.run(None, ort_inputs)
        dets = postprocess(output[0], input_shape, ratio)[0]
        if dets is not None:
            detect_result = detect_watermark(dets, watermark_box)
            return detect_result
        else:
            return False


def postprocess(outputs, img_size, ratio, p6=False):
    grids = []
    expanded_strides = []
    outputs = outputs[0]
    strides = [8, 16, 32] if not p6 else [8, 16, 32, 64]

    hsizes = [img_size[0] // stride for stride in strides]
    wsizes = [img_size[1] // stride for stride in strides]

    for hsize, wsize, stride in zip(hsizes, wsizes, strides):
        xv, yv = np.meshgrid(np.arange(wsize), np.arange(hsize))
        grid = np.stack((xv, yv), 2).reshape(1, -1, 2)
        grids.append(grid)
        shape = grid.shape[:2]
        expanded_strides.append(np.full((*shape, 1), stride))

    grids = np.concatenate(grids, 1)
    expanded_strides = np.concatenate(expanded_strides, 1)
    outputs[..., :2] = (outputs[..., :2] + grids) * expanded_strides
    outputs[..., 2:4] = np.exp(outputs[..., 2:4]) * expanded_strides

    boxes = outputs[:, :4]
    scores = outputs[:, 4:5] * outputs[:, 5:]

    boxes_xyxy = np.ones_like(boxes)
    boxes_xyxy[:, 0] = boxes[:, 0] - boxes[:, 2] / 2.
    boxes_xyxy[:, 1] = boxes[:, 1] - boxes[:, 3] / 2.
    boxes_xyxy[:, 2] = boxes[:, 0] + boxes[:, 2] / 2.
    boxes_xyxy[:, 3] = boxes[:, 1] + boxes[:, 3] / 2.
    boxes_xyxy /= ratio
    dets = multiclass_nms(boxes_xyxy, scores, nms_thr=0.45, score_thr=0.1)

    return dets


def nms(boxes, scores, nms_thr):
    """Single class NMS implemented in Numpy."""
    x1 = boxes[:, 0]
    y1 = boxes[:, 1]
    x2 = boxes[:, 2]
    y2 = boxes[:, 3]

    areas = (x2 - x1 + 1) * (y2 - y1 + 1)
    order = scores.argsort()[::-1]

    keep = []
    while order.size > 0:
        i = order[0]
        keep.append(i)
        xx1 = np.maximum(x1[i], x1[order[1:]])
        yy1 = np.maximum(y1[i], y1[order[1:]])
        xx2 = np.minimum(x2[i], x2[order[1:]])
        yy2 = np.minimum(y2[i], y2[order[1:]])

        w = np.maximum(0.0, xx2 - xx1 + 1)
        h = np.maximum(0.0, yy2 - yy1 + 1)
        inter = w * h
        ovr = inter / (areas[i] + areas[order[1:]] - inter)

        inds = np.where(ovr <= nms_thr)[0]
        order = order[inds + 1]

    return keep


def multiclass_nms_class_agnostic(boxes, scores, nms_thr, score_thr):
    """Multiclass NMS implemented in Numpy. Class-agnostic version."""
    cls_inds = scores.argmax(1)
    cls_scores = scores[np.arange(len(cls_inds)), cls_inds]

    valid_score_mask = cls_scores > score_thr
    if valid_score_mask.sum() == 0:
        return None
    valid_scores = cls_scores[valid_score_mask]
    valid_boxes = boxes[valid_score_mask]
    valid_cls_inds = cls_inds[valid_score_mask]
    keep = nms(valid_boxes, valid_scores, nms_thr)
    if keep:
        dets = np.concatenate(
            [valid_boxes[keep], valid_scores[keep, None], valid_cls_inds[keep, None]], 1
        )
    return dets


def multiclass_nms_class_aware(boxes, scores, nms_thr, score_thr):
    """Multiclass NMS implemented in Numpy. Class-aware version."""
    final_dets = []
    num_classes = scores.shape[1]
    for cls_ind in range(num_classes):
        cls_scores = scores[:, cls_ind]
        valid_score_mask = cls_scores > score_thr
        if valid_score_mask.sum() == 0:
            continue
        else:
            valid_scores = cls_scores[valid_score_mask]
            valid_boxes = boxes[valid_score_mask]
            keep = nms(valid_boxes, valid_scores, nms_thr)
            if len(keep) > 0:
                cls_inds = np.ones((len(keep), 1)) * cls_ind
                dets = np.concatenate(
                    [valid_boxes[keep], valid_scores[keep, None], cls_inds], 1
                )
                final_dets.append(dets)
    if len(final_dets) == 0:
        return None
    return np.concatenate(final_dets, 0)


def multiclass_nms(boxes, scores, nms_thr, score_thr, class_agnostic=True):
    """Multiclass NMS implemented in Numpy"""
    if class_agnostic:
        nms_method = multiclass_nms_class_agnostic
    else:
        nms_method = multiclass_nms_class_aware
    return nms_method(boxes, scores, nms_thr, score_thr)


def detect_watermark(dets, watermark_box, threshold=0.5):
    if dets.size == 0:  # 检查是否为空
        return False
    for box, score, cls in zip(dets[:, :4], dets[:, 4], dets[:, 5]):
        wm_box_coords = watermark_box[:4]
        wm_cls = watermark_box[4]
        if cls == wm_cls:
            ciou = calculate_ciou(box, wm_box_coords)
            if ciou > threshold:
                return True
    return False