#encoding=gbk
import tensorrt as trt
import numpy as np
import os
import cv2
import pycuda.driver as cuda
import pycuda.autoinit
from imutils import paths
from tqdm import tqdm


class HostDeviceMem(object):
    def __init__(self, host_mem, device_mem):
        self.host = host_mem
        self.device = device_mem

    def __str__(self):
        return "Host:\n" + str(self.host) + "\nDevice:\n" + str(self.device)

    def __repr__(self):
        return self.__str__()


class TrtModel:

    def __init__(self, engine_path, max_batch_size=1, dtype=np.float32):

        self.engine_path = engine_path
        self.dtype = dtype
        self.logger = trt.Logger(trt.Logger.WARNING)
        self.runtime = trt.Runtime(self.logger)
        self.engine = self.load_engine(self.runtime, self.engine_path)
        self.max_batch_size = max_batch_size
        self.inputs, self.outputs, self.bindings, self.stream = self.allocate_buffers()
        self.context = self.engine.create_execution_context()

    @staticmethod
    def load_engine(trt_runtime, engine_path):
        trt.init_libnvinfer_plugins(None, "")
        with open(engine_path, 'rb') as f:
            engine_data = f.read()
        engine = trt_runtime.deserialize_cuda_engine(engine_data)
        return engine

    def allocate_buffers(self):

        inputs = []
        outputs = []
        bindings = []
        stream = cuda.Stream()

        for binding in self.engine:
            # size = trt.volume(self.engine.get_binding_shape(binding)) * self.max_batch_size
            #*******
            ssize = self.engine.get_binding_shape(binding)
            ssize[0]=self.max_batch_size
            size=trt.volume(ssize)
            #*******
            host_mem = cuda.pagelocked_empty(size, self.dtype)
            device_mem = cuda.mem_alloc(host_mem.nbytes)

            bindings.append(int(device_mem))

            if self.engine.binding_is_input(binding):
                inputs.append(HostDeviceMem(host_mem, device_mem))
            else:
                outputs.append(HostDeviceMem(host_mem, device_mem))

        return inputs, outputs, bindings, stream

    def __call__(self, x: np.ndarray, batch_size=2):

        x = x.astype(self.dtype)

        np.copyto(self.inputs[0].host, x.ravel())

        for inp in self.inputs:
            cuda.memcpy_htod_async(inp.device, inp.host, self.stream)

        #**********
        origin_inputshape=self.engine.get_binding_shape(0)
        origin_inputshape[0]=batch_size
        self.context.set_binding_shape(0,(origin_inputshape))
        #**********

        self.context.execute_async(batch_size=batch_size, bindings=self.bindings, stream_handle=self.stream.handle)
        for out in self.outputs:
            cuda.memcpy_dtoh_async(out.host, out.device, self.stream)

        self.stream.synchronize()


        return [out.host.reshape(batch_size, -1) for out in self.outputs]


if __name__ == "__main__":

    # 驗(yàn)證模式：fp32,fp16,int8
    val_type='fp16'

#---------------------------------
    path=r'./imgs/'
    trt_engine_path = r'./model/{}.engine'.format(val_type)
    out_path=r'./out/{}'.format(val_type)

    if not os.path.exists(out_path):
        os.makedirs(out_path)

    #均值和方差
    mean = (120, 114, 104)
    std = (70, 69, 73)

    #構(gòu)建模型
    model = TrtModel(trt_engine_path)

    pic_paths = list(paths.list_images(path))
    for pic_path in tqdm(pic_paths):
        name=os.path.basename(pic_path).split('.')[0]
        # 輸入圖像預(yù)處理
        img = cv2.imread(pic_path)
        imgbak = img.copy()
        img = img[:, :, ::-1]
        img = np.array(img).astype(np.float32)  # 注意輸入type一定要np.float32
        img -= mean  # 減均值
        img /= std  # 除方差
        img = np.array([np.transpose(img, (2, 0, 1))])

        #模型推理
        result = model(img, 1)

        # 保存圖像
        img_out=np.reshape(result[0][0],(512,512))
        img_out =img_out.astype('uint8')
        # img_out=img_out*25
        img_out[img_out>0]=255
        cv2.imwrite(os.path.join(out_path,'{}_{}.png'.format(val_type,name)),img_out)