model_watermark_detect/watermark_verify/inference/rcnn_inference.py

"""
定义Faster-RCNN推理流程
"""
import numpy as np
from mindx.sdk import Tensor  # mxVision 中的 Tensor 数据结构
from mindx.sdk import base  # mxVision 推理接口
from PIL import Image
from watermark_verify.utils.utils_bbox import DecodeBox


class FasterRCNNInference:
    def __init__(self, model_path, input_size=(600, 600), num_classes=20, num_iou=0.3, confidence=0.5, swap=(2, 0, 1)):
        """
        初始化Faster-RCNN模型推理流程
        :param model_path: 图像分类模型om文件路径
        :param input_size: 模型输入大小
        :param num_classes: 模型目标检测分类数
        :param num_iou: iou阈值
        :param confidence: 置信度阈值
        :param swap: 变换方式，pytorch需要进行轴变换(默认参数)，tensorflow无需进行轴变换
        """
        self.model_path = model_path
        self.input_size = input_size
        self.swap = swap
        self.num_classes = num_classes
        self.nms_iou = num_iou
        self.confidence = confidence

    def input_processing(self, image_path):
        """
        对输入图片进行预处理
        :param image_path: 图片路径
        :return: 图片经过处理完成的ndarray
        """
        image = Image.open(image_path)
        image_shape = np.array(np.shape(image)[0:2])
        # ---------------------------------------------------------#
        #   在这里将图像转换成RGB图像，防止灰度图在预测时报错。
        #   代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB
        # ---------------------------------------------------------#
        if not (len(np.shape(image)) == 3 and np.shape(image)[2] == 3):
            image = image.convert('RGB')
        image_data = resize_image(image, self.input_size, False)
        image_data = np.array(image_data, dtype='float32')
        image_data = image_data / 255.0
        image_data = np.expand_dims(np.transpose(image_data, self.swap).copy(), 0)
        image_data = image_data.astype('float32')
        return image_data, image_shape

    def predict(self, image_path):
        """
        对单张图片进行推理
        :param image_path: 图片路径
        :return: 推理结果
        """
        image_data, image_shape = self.input_processing(image_path)
        # 使用mindx框架进行om权重文件推理
        base.mx_init()
        model = base.model(modelPath=self.model_path)  # 初始化 base.model 类
        if model is None:
            raise Exception("模型导入失败！请检查model_path和device_id.")
        # 确保img_tensor是正确的输入格式
        input_tensors = Tensor(image_data)  # 将numpy转为转为Tensor类
        scale = Tensor(1.0)
        outputs = model.infer([input_tensors, scale])  # 执行推理
        output = []
        for item in outputs:
            item.to_host()  # 将Tensor数据转移到内存
            item = np.array(item)
            if item.size == 0:
                return False
            output.append(item)
        output = self.output_processing(output, image_shape)
        return output

    def output_processing(self, outputs, image_shape):
        """
        对模型输出进行后处理工作
        :param outputs: 模型原始输出
        :param image_shape: 原始图像大小
        :return: 经过处理完成的模型输出
        """
        # 处理模型预测输出
        roi_cls_locs, roi_scores, rois, _ = outputs
        bbox_util = DecodeBox(self.num_classes)
        results = bbox_util.forward(roi_cls_locs, roi_scores, rois, image_shape, self.input_size,
                                    nms_iou=self.nms_iou, confidence=self.confidence)
        return results


def resize_image(image, size, letterbox_image):
    iw, ih = image.size
    w, h = size
    if letterbox_image:
        scale = min(w / iw, h / ih)
        nw = int(iw * scale)
        nh = int(ih * scale)

        image = image.resize((nw, nh), Image.BICUBIC)
        new_image = Image.new('RGB', size, (128, 128, 128))
        new_image.paste(image, ((w - nw) // 2, (h - nh) // 2))
    else:
        new_image = image.resize((w, h), Image.BICUBIC)
    return new_image