# -*- coding: UTF-8 -*-
import argparse
import copy

import pandas as pd
import pytorch_ssim
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim.lr_scheduler import MultiStepLR
import torchvision.transforms as transforms
from torch.utils.data import DataLoader
from tqdm import tqdm

from data_utils import DatasetFromFolder
from model import Net
from psnrmeter import PSNRMeter


def train(data, target, model, optimizer, loss_fn):
    if torch.cuda.is_available():
        data = data.cuda()
        target = target.cuda()
    pred = model(data)
    loss = loss_fn(pred, target)
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()
    return loss.item()


def valid(val_data, val_target, model):
    if torch.cuda.is_available():
        val_data = val_data.cuda()
        val_target = val_target.cuda()
    val_pred = model(val_data)
    Instance_PSNR = PSNRMeter()
    Instance_PSNR.add(val_pred, val_target)
    psnr = Instance_PSNR.value()
    ssim = pytorch_ssim.ssim(val_pred.cpu(), val_target.cpu()).item()
    return psnr, ssim


if __name__ == "__main__":

    # 这里设置的放大倍数要与预处理时创建的数据集相一致
    parser = argparse.ArgumentParser(description='Train Super Resolution')
    parser.add_argument('--upscale_factor', default=3, type=int, help='super resolution upscale factor')
    parser.add_argument('--num_epochs', default=100, type=int, help='super resolution epochs number')
    opt = parser.parse_args()

    UPSCALE_FACTOR = opt.upscale_factor
    NUM_EPOCHS = opt.num_epochs

    train_set = DatasetFromFolder('data/train', upscale_factor=UPSCALE_FACTOR, input_transform=transforms.ToTensor(),
                                  target_transform=transforms.ToTensor())
    val_set = DatasetFromFolder('data/val', upscale_factor=UPSCALE_FACTOR, input_transform=transforms.ToTensor(),
                                target_transform=transforms.ToTensor())
    train_loader = DataLoader(dataset=train_set, num_workers=4, batch_size=64, shuffle=True)
    val_loader = DataLoader(dataset=val_set, num_workers=4, batch_size=64, shuffle=False)

    model = Net(upscale_factor=UPSCALE_FACTOR)
    criterion = nn.MSELoss()
    if torch.cuda.is_available():
        model = model.cuda()
        criterion = criterion.cuda()

    print('# parameters:', sum(param.numel() for param in model.parameters()))

    optimizer = optim.Adam(model.parameters(), lr=1e-2)
    scheduler = MultiStepLR(optimizer, milestones=[30, 80], gamma=0.1)
    result_loss = []
    result_psnr = []
    result_ssim = []
    best_psnr = 0
    best_ssim = 0
    best_epoch = 1
    best_weight = copy.deepcopy(model.state_dict)
    # 此处模型权重保留最后一次训练结果，避免有时loss会降下来，但指标不一定是最优的情况。
    last_weight = copy.deepcopy(model.state_dict)
    for epoch in range(NUM_EPOCHS):
        model = model.train()
        train_bar = tqdm(train_loader)
        loss_epoch = 0
        trainbatchnum = 0
        for train_data, train_label in train_bar:
            loss_bat = train(train_data, train_label, model, optimizer, criterion)
            loss_epoch += loss_bat
            trainbatchnum += 1

            train_bar.set_description(
                desc='Train[%d/%d] Loss:%.5f' % ((epoch + 1), NUM_EPOCHS, loss_epoch / trainbatchnum))
        model = model.eval()
        with torch.no_grad():
            val_bar = tqdm(val_loader)
            psnr_epoch = 0
            ssim_epoch = 0
            valbatch = 0
            for val_data, val_label in val_bar:
                psnr_batch, ssim_batch = valid(val_data, val_label, model)
                psnr_epoch += psnr_batch
                ssim_epoch += ssim_batch
                valbatch += 1
                val_bar.set_description('psnr: %.5f dB, ssim: %.5f' % (psnr_epoch / valbatch, ssim_epoch / valbatch))
            if (psnr_epoch / valbatch) > best_psnr and (ssim_epoch / valbatch) > best_ssim:
                best_epoch = (epoch + 1)
                best_psnr = (psnr_epoch / valbatch)
                best_ssim = (ssim_epoch / valbatch)
                best_weight = copy.deepcopy(model.state_dict())
        result_loss.append(loss_epoch / trainbatchnum)
        result_psnr.append(psnr_epoch / valbatch)
        result_ssim.append(ssim_epoch / valbatch)
        scheduler.step()
    last_weight = copy.deepcopy(model.state_dict())
    data_frame = pd.DataFrame(
        data={'Loss': result_loss, 'ValidSet_PSNR': result_psnr, 'ValidSet_SSIM': result_ssim},
        index=range(1, NUM_EPOCHS + 1))
    data_frame.to_csv('SRF_' + str(UPSCALE_FACTOR) + '_train_results.csv', index_label='Epoch')
    print("数据保存成功！")
    torch.save(best_weight, "epochs/ESPCN_best({}).pt".format(best_epoch))
    torch.save(last_weight, "epochs/ESPCN_last.pt")
    print("模型保存成功！最佳数据——epoch{}, psnr: {:.5f}dB, ssim: {:.5f}".format(best_epoch, best_psnr, best_ssim))