{"id":297,"date":"2025-05-17T19:25:39","date_gmt":"2025-05-17T11:25:39","guid":{"rendered":"http:\/\/www.s7arry.top\/?p=297"},"modified":"2025-05-17T19:31:33","modified_gmt":"2025-05-17T11:31:33","slug":"pytorch2","status":"publish","type":"post","link":"https:\/\/www.s7arry.top\/index.php\/2025\/05\/17\/pytorch2\/","title":{"rendered":"\u51c6\u725b\u9a6c\u7684Pytorch\u5b66\u4e60\uff082\uff09\u6570\u636e\u5904\u7406"},"content":{"rendered":"\n

DataLoader\u4e0eDataSet<\/h1>\n\n\n\n

\u4eba\u6c11\u5e01\u4e8c\u5206\u7c7b\u95ee\u9898<\/h2>\n\n\n\n

\u5047\u5982\u6211\u4eec\u8981\u5bf9100\u5143\u548c1\u5143\u4eba\u6c11\u5e01\u8fdb\u884c\u4e8c\u5206\u7c7b\uff0c\u90a3\u4e48\u6211\u4eec\u9700\u8981\u8f93\u5165\u56fe\u7247\u4f5c\u4e3a\u6570\u636e\uff0c\u8fd9\u4e0e\u4e4b\u524d\u8f93\u5165\u7684\u6570\u636e\u5f62\u5f0f\u5b8c\u5168\u4e0d\u540c\uff0c\u6211\u4eec\u5e94\u8be5\u5982\u4f55\u5b9e\u73b0\u5462\uff1f<\/p>\n\n\n\n

\"\"<\/div><\/figure>\n\n\n\n

\u5728Pytorch\u4e2d\uff0c\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86DataLoader\u6765\u8fdb\u884c\u56fe\u7247\u6570\u636e\u7684\u8bfb\u53d6<\/p>\n\n\n\n

\"\"<\/div><\/figure>\n\n\n\n

Sampler\uff1a\u7ed9\u6570\u636e\u751f\u6210\u7d22\u5f15<\/p>\n\n\n\n

DataSet\uff1a\u6839\u636e\u7d22\u5f15\u6765\u8fdb\u884c\u8bfb\u53d6<\/p>\n\n\n\n

\u63a5\u4e0b\u6765\u8be6\u7ec6\u4ecb\u7ecdDataLoader\u548cDataSet<\/p>\n\n\n\n

DataLoader<\/h2>\n\n\n\n

DataLoader\u7684\u529f\u80fd\u5c31\u662f\u6784\u5efa\u53ef\u8fed\u4ee3\u7684\u6570\u636e\u88c5\u8f7d\u5668\uff0c\u5176\u4e2d\u6709\u5f88\u591a\u7684\u53c2\u6570\u53ef\u4ee5\u8c03\u6574\uff0c\u8fd9\u91cc\u4ecb\u7ecd5\u4e2a\u5e38\u7528\u7684\u53c2\u6570<\/p>\n\n\n\n

\"\"<\/div><\/figure>\n\n\n\n

\u8be6\u7ec6\u89e3\u91ca\u4e00\u4e0b\u51e0\u4e2a\u540d\u8bcd\uff1a<\/p>\n\n\n\n

    \n
  • Epoch\uff1a\u6240\u6709\u8bad\u7ec3\u6837\u672c\u90fd\u5df2\u8f93\u5165\u5230\u6a21\u578b\u4e2d\uff0c\u79f0\u4e3a\u4e00\u4e2aEpoch<\/li>\n\n\n\n
  • Iteration\uff1a\u4e00\u6279\u6837\u672c\u8f93\u5165\u5230\u6a21\u578b\u4e2d\uff0c\u79f0\u4e4b\u4e3a\u4e00\u4e2aIteration<\/li>\n\n\n\n
  • Batchsize\uff1a\u6279\u5927\u5c0f\uff0c\u51b3\u5b9a\u4e00\u4e2aEpoch\u6709\u591a\u5c11\u4e2aIteration<\/li>\n<\/ul>\n\n\n\n

    \u4e3e\u4f8b\u8bf4\u660e\uff1a<\/p>\n\n\n\n

    \u5982\u679c\u6837\u672c\u603b\u6570\u4e3a80\uff0cBatchsize\u4e3a8\uff0c\u90a3\u4e481 Epoch = 10 Iteration<\/p>\n\n\n\n

    \u90a3\u5047\u5982\u6837\u672c\u603b\u6570\u548cBatchsize\u4e0d\u80fd\u6574\u9664\u600e\u4e48\u5904\u7406\u5462\uff1f<\/p>\n\n\n\n

    \u8fd9\u5c31\u4e0edrop_last\u53c2\u6570\u6709\u5173\uff0c\u662f\u5426\u820d\u5f03\u6700\u540e\u4e00\u6279\u6570\u636e<\/p>\n\n\n\n

    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n

    DataSet<\/h2>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n

    \u4eba\u6c11\u5e01\u4e8c\u5206\u7c7b\u9879\u76ee<\/h2>\n\n\n\n

    \u6211\u4eec\u56de\u5230\u4e0a\u9762\u7684\u95ee\u9898\u4e2d\uff0c\u901a\u8fc7\u9879\u76ee\u4ee3\u7801\u6765\u5b66\u4e60DataLoader\u548cDataSet\u7684\u8fd0\u884c\u903b\u8f91\uff0c\u4ee5\u53ca\u6570\u636e\u8bfb\u53d6\u7684\u4e09\u4e2a\u95ee\u9898\u3002<\/p>\n\n\n\n

    \"\"<\/div><\/figure>\n\n\n\n

    \u5212\u5206\u8bad\u7ec3\u96c6\u3001\u9a8c\u8bc1\u96c6\u548c\u6d4b\u8bd5\u96c6<\/h3>\n\n\n\n

    \u9996\u5148\u6211\u4eec\u6253\u5f00split_dataset.py\uff0c\u8fd9\u5757\u4ee3\u7801\u7684\u4f5c\u7528\u5c31\u662f\u5e2e\u6211\u4eec\u5c06\u6570\u636e\u4eba\u6c11\u5e01\u56fe\u7247\u8fdb\u884c\u6570\u636e\u5212\u5206\u3002<\/p>\n\n\n\n

    # -*- coding: utf-8 -*-\n\"\"\"\n# @file name  : 1_split_dataset.py\n# @author     : tingsongyu\n# @date       : 2019-09-07 10:08:00\n# @brief      : \u5c06\u6570\u636e\u96c6\u5212\u5206\u4e3a\u8bad\u7ec3\u96c6\uff0c\u9a8c\u8bc1\u96c6\uff0c\u6d4b\u8bd5\u96c6\n\"\"\"\n\nimport os\nimport random\nimport shutil\n\n\ndef makedir(new_dir):\n    if not os.path.exists(new_dir):\n        os.makedirs(new_dir)\n\n\nif __name__ == '__main__':\n\n    random.seed(1)\n\ndataset_dir = os.path.join(\"..\", \"..\", \"data\", \"RMB_data\")\nsplit_dir = os.path.join(\"..\", \"..\", \"data\", \"rmb_split\")\ntrain_dir = os.path.join(split_dir, \"train\")\nvalid_dir = os.path.join(split_dir, \"valid\")\ntest_dir = os.path.join(split_dir, \"test\")\n\ntrain_pct = 0.8\nvalid_pct = 0.1\ntest_pct = 0.1\n\nfor root, dirs, files in os.walk(dataset_dir):\n    for sub_dir in dirs:\n\n        imgs = os.listdir(os.path.join(root, sub_dir))\nimgs = list(filter(lambda x: x.endswith('.jpg'), imgs))\nrandom.shuffle(imgs)\nimg_count = len(imgs)\n\ntrain_point = int(img_count * train_pct)\nvalid_point = int(img_count * (train_pct + valid_pct))\n\nfor i in range(img_count):\n    if i < train_point:\n        out_dir = os.path.join(train_dir, sub_dir)\n    elif i < valid_point:\n        out_dir = os.path.join(valid_dir, sub_dir)\n    else:\n        out_dir = os.path.join(test_dir, sub_dir)\n\nmakedir(out_dir)\n\ntarget_path = os.path.join(out_dir, imgs[i])\nsrc_path = os.path.join(dataset_dir, sub_dir, imgs[i])\n\nshutil.copy(src_path, target_path)\n\nprint('Class:{}, train:{}, valid:{}, test:{}'.format(sub_dir, train_point, valid_point-train_point,\n                                                     img_count-valid_point))\n<\/code><\/pre>\n\n\n\n
    \u8fd0\u884c\u540e\uff0c\u6211\u4eec\u53ef\u4ee5\u5f97\u5230\u5212\u5206\u597d\u7684\u6570\u636e\u6587\u4ef6<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n
    \u6a21\u578b\u8bad\u7ec3<\/h3>\n\n\n\n
    \u6a21\u578b\u8bad\u7ec3\u4e3b\u8981\u5305\u542b\u4e86\u4e94\u4e2a\u90e8\u5206\uff1a\u6570\u636e\u3001\u6a21\u578b\u3001\u635f\u5931\u51fd\u6570\u3001\u4f18\u5316\u5668\u548c\u8bad\u7ec3\u3002<\/p>\n\n\n\n
    # -*- coding: utf-8 -*-\n\"\"\"\n# @file name  : train_lenet.py\n# @author     : tingsongyu\n# @date       : 2019-09-07 10:08:00\n# @brief      : \u4eba\u6c11\u5e01\u5206\u7c7b\u6a21\u578b\u8bad\u7ec3\n\"\"\"\nimport sys\nsys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), \"..\/..\/\")))\nimport os\nimport random\nimport numpy as np\nimport torch\nimport torch.nn as nn\nfrom torch.utils.data import DataLoader\nimport torchvision.transforms as transforms\nimport torch.optim as optim\nfrom matplotlib import pyplot as plt\nfrom model.lenet import LeNet\nfrom tools.my_dataset import RMBDataset\n\n\ndef set_seed(seed=1):\n    random.seed(seed)\n    np.random.seed(seed)\n    torch.manual_seed(seed)\n    torch.cuda.manual_seed(seed)\n\n\nset_seed()  # \u8bbe\u7f6e\u968f\u673a\u79cd\u5b50\nrmb_label = {\"1\": 0, \"100\": 1}\n\n# \u53c2\u6570\u8bbe\u7f6e\nMAX_EPOCH = 10\nBATCH_SIZE = 16\nLR = 0.01\nlog_interval = 10\nval_interval = 1\n\n# ============================ step 1\/5 \u6570\u636e ============================\n\nsplit_dir = os.path.join(\"..\", \"..\", \"data\", \"rmb_split\")\ntrain_dir = os.path.join(split_dir, \"train\")\nvalid_dir = os.path.join(split_dir, \"valid\")\n\nnorm_mean = [0.485, 0.456, 0.406]\nnorm_std = [0.229, 0.224, 0.225]\n\ntrain_transform = transforms.Compose([\n    transforms.Resize((32, 32)),\n    transforms.RandomCrop(32, padding=4),\n    transforms.ToTensor(),\n    transforms.Normalize(norm_mean, norm_std),\n])\n\nvalid_transform = transforms.Compose([\n    transforms.Resize((32, 32)),\n    transforms.ToTensor(),\n    transforms.Normalize(norm_mean, norm_std),\n])\n\n# \u6784\u5efaMyDataset\u5b9e\u4f8b\ntrain_data = RMBDataset(data_dir=train_dir, transform=train_transform)\nvalid_data = RMBDataset(data_dir=valid_dir, transform=valid_transform)\n\n# \u6784\u5efaDataLoder\ntrain_loader = DataLoader(dataset=train_data, batch_size=BATCH_SIZE, shuffle=True)\nvalid_loader = DataLoader(dataset=valid_data, batch_size=BATCH_SIZE)\n\n# ============================ step 2\/5 \u6a21\u578b ============================\n\nnet = LeNet(classes=2)\nnet.initialize_weights()\n\n# ============================ step 3\/5 \u635f\u5931\u51fd\u6570 ============================\ncriterion = nn.CrossEntropyLoss()                                                   # \u9009\u62e9\u635f\u5931\u51fd\u6570\n\n# ============================ step 4\/5 \u4f18\u5316\u5668 ============================\noptimizer = optim.SGD(net.parameters(), lr=LR, momentum=0.9)                        # \u9009\u62e9\u4f18\u5316\u5668\nscheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=10, gamma=0.1)     # \u8bbe\u7f6e\u5b66\u4e60\u7387\u4e0b\u964d\u7b56\u7565\n\n# ============================ step 5\/5 \u8bad\u7ec3 ============================\ntrain_curve = list()\nvalid_curve = list()\n\nfor epoch in range(MAX_EPOCH):\n\n    loss_mean = 0.\n    correct = 0.\n    total = 0.\n\n    net.train()\n    for i, data in enumerate(train_loader):\n\n        # forward\n        inputs, labels = data\n        outputs = net(inputs)\n\n        # backward\n        optimizer.zero_grad()\n        loss = criterion(outputs, labels)\n        loss.backward()\n\n        # update weights\n        optimizer.step()\n\n        # \u7edf\u8ba1\u5206\u7c7b\u60c5\u51b5\n        _, predicted = torch.max(outputs.data, 1)\n        total += labels.size(0)\n        correct += (predicted == labels).squeeze().sum().numpy()\n\n        # \u6253\u5370\u8bad\u7ec3\u4fe1\u606f\n        loss_mean += loss.item()\n        train_curve.append(loss.item())\n        if (i+1) % log_interval == 0:\n            loss_mean = loss_mean \/ log_interval\n            print(\"Training:Epoch[{:0>3}\/{:0>3}] Iteration[{:0>3}\/{:0>3}] Loss: {:.4f} Acc:{:.2%}\".format(\n                epoch, MAX_EPOCH, i+1, len(train_loader), loss_mean, correct \/ total))\n            loss_mean = 0.\n\n    scheduler.step()  # \u66f4\u65b0\u5b66\u4e60\u7387\n\n    # validate the model\n    if (epoch+1) % val_interval == 0:\n\n        correct_val = 0.\n        total_val = 0.\n        loss_val = 0.\n        net.eval()\n        with torch.no_grad():\n            for j, data in enumerate(valid_loader):\n                inputs, labels = data\n                outputs = net(inputs)\n                loss = criterion(outputs, labels)\n\n                _, predicted = torch.max(outputs.data, 1)\n                total_val += labels.size(0)\n                correct_val += (predicted == labels).squeeze().sum().numpy()\n\n                loss_val += loss.item()\n\n            loss_val_epoch = loss_val \/ len(valid_loader)\n            valid_curve.append(loss_val_epoch)\n            # valid_curve.append(loss.item())    # 20191022\u6539\uff0c\u8bb0\u5f55\u6574\u4e2aepoch\u6837\u672c\u7684loss\uff0c\u6ce8\u610f\u8981\u53d6\u5e73\u5747\n            print(\"Valid:\\t Epoch[{:0>3}\/{:0>3}] Iteration[{:0>3}\/{:0>3}] Loss: {:.4f} Acc:{:.2%}\".format(\n                epoch, MAX_EPOCH, j+1, len(valid_loader), loss_val_epoch, correct_val \/ total_val))\n\n\ntrain_x = range(len(train_curve))\ntrain_y = train_curve\n\ntrain_iters = len(train_loader)\nvalid_x = np.arange(1, len(valid_curve)+1) * train_iters*val_interval # \u7531\u4e8evalid\u4e2d\u8bb0\u5f55\u7684\u662fepochloss\uff0c\u9700\u8981\u5bf9\u8bb0\u5f55\u70b9\u8fdb\u884c\u8f6c\u6362\u5230iterations\nvalid_y = valid_curve\n\nplt.plot(train_x, train_y, label='Train')\nplt.plot(valid_x, valid_y, label='Valid')\n\nplt.legend(loc='upper right')\nplt.ylabel('loss value')\nplt.xlabel('Iteration')\nplt.show()\n\n# ============================ inference ============================\n\nBASE_DIR = os.path.dirname(os.path.abspath(__file__))\ntest_dir = os.path.join(BASE_DIR, \"test_data\")\n\ntest_data = RMBDataset(data_dir=test_dir, transform=valid_transform)\nvalid_loader = DataLoader(dataset=test_data, batch_size=1)\n\nfor i, data in enumerate(valid_loader):\n    # forward\n    inputs, labels = data\n    outputs = net(inputs)\n    _, predicted = torch.max(outputs.data, 1)\n\n    rmb = 1 if predicted.numpy()[0] == 0 else 100\n    print(\"\u6a21\u578b\u83b7\u5f97{}\u5143\".format(rmb))<\/code><\/pre>\n\n\n\n
    \u53ef\u4ee5\u5148\u8fd0\u884c\u4e0b\uff0c\u770b\u770b\u6548\u679c\uff1a<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \u7ecf\u8fc7\u5341\u8f6e\u7684\u8bad\u7ec3\uff0c\u6700\u540e\u753b\u51fa\u4e86\u4e00\u4e2aloss\u7684\u56fe<\/p>\n\n\n\n
    \u63a5\u4e0b\u6765\u8be6\u7ec6\u89e3\u91ca\u6570\u636e\u90e8\u5206<\/p>\n\n\n\n
    \n\n\n\n
    \u6570\u636e\u90e8\u5206<\/h3>\n\n\n\n
    \u9996\u5148\u662f\u5b9a\u4e49\u6570\u636e\u7684\u8def\u5f84\uff0c\u5c06\u8bad\u7ec3\u96c6\u548c\u9a8c\u8bc1\u96c6\u7684\u8def\u5f84\u5b9a\u4e49\u597d<\/p>\n\n\n\n
    split_dir = os.path.join(\"..\", \"..\", \"data\", \"rmb_split\")\ntrain_dir = os.path.join(split_dir, \"train\")\nvalid_dir = os.path.join(split_dir, \"valid\")<\/code><\/pre>\n\n\n\n
    \u4e0b\u4e00\u6b65\u5bf9\u6570\u636e\u8fdb\u884c\u9884\u5904\u7406\uff0c\u5bf9\u6570\u636e\u8fdb\u884c\u4e00\u4e9b\u53d8\u6362\uff0c\u5c06\u5176\u8f6c\u6362\u4e3a\u5f20\u91cf<\/p>\n\n\n\n
    train_transform = transforms.Compose([\n    transforms.Resize((32, 32)),\n    transforms.RandomCrop(32, padding=4),\n    transforms.ToTensor(),\n    transforms.Normalize(norm_mean, norm_std),\n])\n\nvalid_transform = transforms.Compose([\n    transforms.Resize((32, 32)),\n    transforms.ToTensor(),\n    transforms.Normalize(norm_mean, norm_std),\n])<\/code><\/pre>\n\n\n\n
    \u63a5\u4e0b\u6765\u5c31\u662f\u91cd\u70b9\uff0c\u6784\u5efaDataSet\u548cDataLoader\u5b9e\u4f8b<\/p>\n\n\n\n
    \u6784\u5efaDataSet\u5b9e\u4f8b\u4e3b\u8981\u6709\u4e24\u4e2a\u91cd\u8981\u53c2\u6570\uff1a\u6570\u636e\u8def\u5f84\uff08\u4ece\u54ea\u91cc\u8bfb\u53d6\u6570\u636e\uff09\uff0c\u6570\u636e\u9884\u5904\u7406<\/p>\n\n\n\n
    # \u6784\u5efaMyDataset\u5b9e\u4f8b\ntrain_data = RMBDataset(data_dir=train_dir, transform=train_transform)\nvalid_data = RMBDataset(data_dir=valid_dir, transform=valid_transform)<\/code><\/pre>\n\n\n\n
    \u4e0a\u6587\u4ecb\u7ecd\u5230\uff0cDataSet\u8fd9\u4e2a\u7c7b\uff0c\u662f\u9700\u8981\u6211\u4eec\u81ea\u5b9a\u4e49\u53bb\u5199\u7684\uff0c\u8fd9\u91cc\u8be6\u7ec6\u89e3\u91ca\u4e00\u4e0b\u8fd9\u4e2a\u7c7b\u662f\u5982\u4f55\u5199\u7684\uff1a<\/p>\n\n\n\n
    a. init\uff0c\u521d\u59cb\u5316\u8fd9\u91cc\u5c31\u4e0d\u591a\u89e3\u91ca\u4e86<\/p>\n\n\n\n
    def __init__(self, data_dir, transform=None):\n        \"\"\"\n        rmb\u9762\u989d\u5206\u7c7b\u4efb\u52a1\u7684Dataset\n        :param data_dir: str, \u6570\u636e\u96c6\u6240\u5728\u8def\u5f84\n        :param transform: torch.transform\uff0c\u6570\u636e\u9884\u5904\u7406\n        \"\"\"\n        self.label_name = {\"1\": 0, \"100\": 1}\n        self.data_info = self.get_img_info(data_dir)  # data_info\u5b58\u50a8\u6240\u6709\u56fe\u7247\u8def\u5f84\u548c\u6807\u7b7e\uff0c\u5728DataLoader\u4e2d\u901a\u8fc7index\u8bfb\u53d6\u6837\u672c\n        self.transform = transform<\/code><\/pre>\n\n\n\n
    b. getitem\uff0c\u8fd9\u4e2a\u662f\u7c7b\u4e2d\u7684\u6838\u5fc3\uff0c\u4e3b\u8981\u662f\u6839\u636eindex\u7d22\u5f15\u6765\u83b7\u53d6\u56fe\u7247\u548c\u6807\u7b7e<\/p>\n\n\n\n
    def __getitem__(self, index):\n        path_img, label = self.data_info[index]\n        img = Image.open(path_img).convert('RGB')     # 0~255\n\n        if self.transform is not None:\n            img = self.transform(img)   # \u5728\u8fd9\u91cc\u505atransform\uff0c\u8f6c\u4e3atensor\u7b49\u7b49\n\n        return img, label<\/code><\/pre>\n\n\n\n
    c. length\uff0c\u8fd4\u56de\u6570\u636e\u7684\u957f\u5ea6<\/p>\n\n\n\n
        def __len__(self):\n        return len(self.data_info)<\/code><\/pre>\n\n\n\n
    d. get_img_info\uff0c\u81ea\u5b9a\u5199\u7684\u65b9\u6cd5\uff0c\u6839\u636e\u8def\u5f84\u6765\u83b7\u53d6\u6570\u636e\u7684\u4fe1\u606f<\/p>\n\n\n\n
    def get_img_info(data_dir):\n        data_info = list()\n        for root, dirs, _ in os.walk(data_dir):\n            # \u904d\u5386\u7c7b\u522b\n            for sub_dir in dirs:\n                img_names = os.listdir(os.path.join(root, sub_dir))\n                img_names = list(filter(lambda x: x.endswith('.jpg'), img_names))\n\n                # \u904d\u5386\u56fe\u7247\n                for i in range(len(img_names)):\n                    img_name = img_names[i]\n                    path_img = os.path.join(root, sub_dir, img_name)\n                    label = rmb_label[sub_dir]\n                    data_info.append((path_img, int(label)))\n\n        return data_info<\/code><\/pre>\n\n\n\n
    \n\n\n\n
    \u56de\u5230\u539f\u6765\u7684\u4ee3\u7801\u4e2d\uff0c\u63a5\u4e0b\u6765\u6784\u5efaDataLoader\u5b9e\u4f8b\uff0c\u4f20\u5165DataSet\u5b9e\u4f8b\uff0c\u8bbe\u7f6ebatchsize\uff0cshuffle\u8868\u793a\u662f\u5426\u4e71\u5e8f<\/p>\n\n\n\n
    # \u6784\u5efaDataLoder\ntrain_loader = DataLoader(dataset=train_data, batch_size=BATCH_SIZE, shuffle=True)\nvalid_loader = DataLoader(dataset=valid_data, batch_size=BATCH_SIZE)<\/code><\/pre>\n\n\n\n
    \n\n\n\n
    \u6570\u636e\u8bfb\u53d6<\/h3>\n\n\n\n
    \u6211\u4eec\u4ece\u6a21\u578b\u8bad\u7ec3\u90e8\u5206\uff0c\u901a\u8fc7\u65ad\u70b9\u8c03\u8bd5\uff0c\u8be6\u7ec6\u770b\u4e00\u4e0b\u662f\u5982\u4f55\u8fdb\u884c\u6570\u636e\u8bfb\u53d6\u7684<\/p>\n\n\n\n
    for i, data in enumerate(train_loader):<\/code><\/pre>\n\n\n\n
    \u5148\u8df3\u8f6c\u5230dataloader.py\uff0c\u9009\u62e9\u5355\u8fdb\u7a0b\u8bfb\u53d6\u8fd8\u662f\u591a\u8fdb\u7a0b\u8bfb\u53d6<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \u4ee5\u5355\u8fdb\u7a0b\u4e3a\u4f8b\uff0c\u968f\u540e\u8df3\u8f6c\u5230\u5355\u8fdb\u7a0b\u7c7b\u4e2d\u8fd9\u4e2a\u51fd\u6570\uff0c\u5b83\u4f1a\u83b7\u53d6index\u548cdata<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n
    \u8fdb\u5165next_index\u8fd9\u4e2a\u51fd\u6570\u4e2d\uff0c\u80fd\u770bsampler\u91c7\u6837\u5668<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \u6211\u4eec\u518d\u8fdb\u5165sampler.py\u7684sampler_iter\u4e2d\uff0c\u8fd9\u4e2a\u51fd\u6570\u5c31\u662f\u544a\u8bc9\u6211\u4eec\u6bcf\u4e2abatch\u83b7\u53d6\u90a3\u4e9b\u6570\u636e<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n
    \u83b7\u53d6index\u540e\uff0c\u6211\u4eec\u56de\u5230\u8fd9\u91cc\uff0c\u63a5\u4e0b\u6765\u770b\u770bdata\u7684\u83b7\u53d6<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \u8fdb\u5165\u5230fetch.py\u4e2d\uff0c\u8fd9\u91cc\u8c03\u7528\u4e86dataset\u7c7b\uff0c\u4f20\u5165index\uff0c\u6839\u636e\u6211\u4eec\u81ea\u5b9a\u4e49\u7684getitem\u51fd\u6570\u6765\u8fd4\u56dedata<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \u6700\u540e\u8c03\u7528\u4e86collate_fn\uff0c\u8fd9\u662f\u4e2a\u6570\u636e\u6574\u7406\u5668\uff0c\u5c06\u6570\u636e\u6574\u7406\u4e3alist\u7684\u5f62\u5f0f\uff0c\u8fd4\u56de\u3002<\/p>\n\n\n\n
    \n\n\n\n
    \u603b\u7ed3<\/h2>\n\n\n\n
    \u8be6\u7ec6\u770b\u4e86\u4ee3\u7801\u4ee5\u540e\uff0c\u6211\u4eec\u6574\u7406\u4e00\u4e0b\u6574\u4e2a\u6570\u636e\u8bfb\u53d6\u7684\u8fc7\u7a0b<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \u7528\u6d41\u7a0b\u56fe\u6765\u8868\u793a\uff1a<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n
    Transforms \u56fe\u50cf\u9884\u5904\u7406<\/h1>\n\n\n\n
    Transforms<\/h2>\n\n\n\n
    \u5728\u4e4b\u524d\u5b89\u88c5\u5305\u7684\u65f6\u5019\uff0c\u6211\u4eec\u5b89\u88c5\u4e86torchvision\u5305\uff0c\u4e5f\u5c31\u662f\u8ba1\u7b97\u89c6\u89c9\u5de5\u5177\u5305\uff0c\u5b83\u5305\u542b\u4e86\u4e00\u7cfb\u5217\u7684\u5de5\u5177<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n
    \u8fd9\u91cc\u4e3b\u8981\u4ecb\u7ecdtransforms<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \u6211\u4eec\u53ef\u4ee5\u770b\u4e0a\u4e00\u8282RMB\u5206\u7c7b\u95ee\u9898\u4e2d\u7684\u9884\u5904\u7406\u90e8\u5206\uff1a<\/p>\n\n\n\n
    train_transform = transforms.Compose([\n    transforms.Resize((32, 32)),\n    transforms.RandomCrop(32, padding=4),\n    transforms.ToTensor(),\n    transforms.Normalize(norm_mean, norm_std),\n])\n\nvalid_transform = transforms.Compose([\n    transforms.Resize((32, 32)),\n    transforms.ToTensor(),\n    transforms.Normalize(norm_mean, norm_std),\n])<\/code><\/pre>\n\n\n\n
    transforms.Compose\uff1a\u662f\u5bf9\u4e00\u7cfb\u5217\u53d8\u6362\u7684\u7ec4\u5408<\/p>\n\n\n\n
    transforms.Resize\uff1a\u5bf9\u56fe\u50cf\u8fdb\u884c\u7f29\u653e<\/p>\n\n\n\n
    transforms.RandomCrop\uff1a\u968f\u673a\u88c1\u526a<\/p>\n\n\n\n
    transforms.ToTensor\uff1a\u8f6c\u6210\u5f20\u91cf<\/p>\n\n\n\n
    transforms.Normalize\uff1a\u6807\u51c6\u5316<\/p>\n\n\n\n
    \n\n\n\n
    \u5728\u4e0a\u4e00\u8282\u4e2d\u8be6\u7ec6\u8bf4\u660e\u4e86\u6570\u636e\u8bfb\u53d6\u4ee3\u7801\u7ec6\u8282\uff0c\u8fd9\u91cc\u4e5f\u4e0e\u4e0a\u9762\u4e00\u6837\uff0c\u8be6\u7ec6\u89e3\u91ca\u4e00\u4e0btransforms\u7684\u7ec6\u8282<\/p>\n\n\n\n
    \u4ece\u81ea\u5b9a\u4e49\u7684dataset\u7c7bgetitem\u51fd\u6570\u5f00\u59cb\uff0c\u8c03\u7528\u4e86transform<\/p>\n\n\n\n
    def __getitem__(self, index):\n        path_img, label = self.data_info[index]\n        img = Image.open(path_img).convert('RGB')     # 0~255\n\n        if self.transform is not None:\n            img = self.transform(img)   # \u5728\u8fd9\u91cc\u505atransform\uff0c\u8f6c\u4e3atensor\u7b49\u7b49\n\n        return img, label<\/code><\/pre>\n\n\n\n
    \u968f\u540e\u8df3\u8f6c\u5230transforms.py\uff0ccall\u51fd\u6570\uff0c\u4f9d\u6b21\u6709\u5e8f\u7684\u8fdb\u884ctransforms.Compose\u4e2d\u9884\u8bbe\u7684\u64cd\u4f5c<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n
    \u968f\u540e\u6211\u4eec\u53ef\u4ee5\u628atransform\u52a0\u5165\u5230\u6d41\u7a0b\u56fe\u4e2d<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n
    transforms.Normalize<\/h2>\n\n\n\n
    \u5c06\u5747\u503c\u53d8\u4e3a0\uff0c\u6807\u51c6\u5dee\u53d8\u4e3a1<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n
    normalize\u7684\u8be6\u7ec6\u5b9e\u73b0\uff1a<\/p>\n\n\n\n
    \u5148\u5224\u65ad\u662f\u5426\u4e3a\u5f20\u91cf\uff0c\u4e0b\u4e00\u6b65\u662f\u5426\u539f\u5730\u64cd\u4f5c\uff0c\u5426\u5219\u514b\u9686\u4e00\u4efd\u3002<\/p>\n\n\n\n
    \u5c06\u5747\u503c\u548c\u6807\u51c6\u5dee\u8f6c\u6362\u4e3a\u5f20\u91cf\u7684\u5f62\u5f0f\uff0c\u6839\u636e\u516c\u5f0f\u8fdb\u884c\u8ba1\u7b97\u3002<\/p>\n\n\n\n
    \u8fd4\u56de\u5f20\u91cf\u3002<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n
    Transforms \u56fe\u50cf\u589e\u5f3a\uff08\u4e00\uff09<\/h1>\n\n\n\n
    \u6570\u636e\u589e\u5f3a<\/h2>\n\n\n\n
    \u6570\u636e\u589e\u5f3a\u53c8\u79f0\u4e3a\u6570\u636e\u589e\u5e7f\uff0c\u6570\u636e\u6269\u589e\uff0c\u5b83\u662f\u5bf9\u8bad\u7ec3\u96c6\u8fdb\u884c\u53d8\u6362\uff0c\u4f7f\u8bad\u7ec3\u96c6\u66f4\u4e30\u5bcc\uff0c\u4ece\u800c\u8ba9\u6a21\u578b\u66f4\u5177\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n\n\n\n
    \u4ee5\u8fd9\u5f20\u56fe\u4e3a\u4f8b\uff0c\u901a\u8fc7\u4e00\u7cfb\u5217\u65b9\u5f0f\u5c06\u56fe\u7247\u53d8\u4e3a\u66f4\u591a\u5f20\u56fe\u7247\uff0c\u4f9b\u6a21\u578b\u8bad\u7ec3\uff0c\u8ba9\u6a21\u578b\u5177\u6709\u66f4\u5f3a\u7684\u6cdb\u5316\u80fd\u529b<\/p>\n\n\n\n
    \"\"<\/div><\/figure>\n\n\n\n
    \n\n\n\n
    \u88c1\u526a Crop<\/h2>\n\n\n\n
    1. transforms.CenterCrop<\/h3>\n\n\n\n
    \u529f\u80fd\uff1a\u4ece\u56fe\u50cf\u4e2d\u5fc3\u88c1\u526a\u56fe\u7247<\/p>\n\n\n\n
      \n
    • size\uff1a\u6240\u9700\u88c1\u526a\u56fe\u7247\u5c3a\u5bf8<\/li>\n<\/ul>\n\n\n\n
      \"\"<\/div><\/figure>\n\n\n\n
      2. transforms.RandomCrop<\/h3>\n\n\n\n
      \u529f\u80fd\uff1a\u4ece\u56fe\u7247\u4e2d\u968f\u673a\u88c1\u526a\u51fa\u5c3a\u5bf8\u4e3asize\u7684\u56fe\u7247<\/p>\n\n\n\n
      \"\"<\/div><\/figure>\n\n\n\n
        \n
      • size\uff1a\u6240\u9700\u88c1\u526a\u56fe\u7247\u5c3a\u5bf8<\/li>\n\n\n\n
      • padding\uff1a\u8bbe\u7f6e\u586b\u5145\u5927\u5c0f\n
          \n
        • \u5f53\u4e3aa\u65f6\uff0c\u4e0a\u4e0b\u5de6\u53f3\u5747\u586b\u5145a\u4e2a\u50cf\u7d20<\/li>\n\n\n\n
        • \u5f53\u4e3a\uff08a,b\uff09\u65f6\uff0c\u4e0a\u4e0b\u586b\u5145b\u4e2a\u50cf\u7d20\uff0c\u5de6\u53f3\u586b\u5145a\u4e2a\u50cf\u7d20<\/li>\n\n\n\n
        • \u5f53\u4e3a\uff08a,b,c,d\uff09\u65f6\uff0c\u5de6\u3001\u4e0a\u3001\u53f3\u3001\u4e0b\u5206\u522b\u586b\u5145a,b,c,d<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n
            \n
          • pad_if_need\uff1a\u82e5\u56fe\u50cf\u5c0f\u4e8e\u8bbe\u5b9asize\uff0c\u5219\u586b\u5145<\/li>\n\n\n\n
          • padding_mode\uff1a\u586b\u5145\u6a21\u5f0f\uff0c\u67094\u79cd\u6a21\u5f0f<\/li>\n<\/ul>\n\n\n\n
              \n
            • constant\uff1a\u50cf\u7d20\u503c\u7531fill\u8bbe\u5b9a<\/li>\n\n\n\n
            • edge\uff1a\u50cf\u7d20\u503c\u7531\u56fe\u50cf\u8fb9\u7f18\u50cf\u7d20\u51b3\u5b9a<\/li>\n\n\n\n
            • reflect\uff1a\u955c\u50cf\u586b\u5145\uff0c\u6700\u540e\u4e00\u4e2a\u50cf\u7d20\u4e0d\u955c\u50cf\uff0ceg:[1,2,3,4]\u2192[3,2,1,2,3,4,3,2]<\/li>\n\n\n\n
            • symmetric\uff1a\u955c\u50cf\u586b\u5145\uff0c\u6700\u540e\u4e00\u4e2a\u50cf\u7d20\u955c\u50cf\uff0ceg:[1,2,3,4]\u2192[2,1,1,2,3,4,4,3]<\/li>\n\n\n\n
            • fill\uff1aconstant\u65f6\uff0c\u8bbe\u7f6e\u586b\u5145\u7684\u50cf\u7d20\u503c<\/li>\n<\/ul>\n\n\n\n
              3. transforms.RandomResizedCrop<\/h3>\n\n\n\n
              \u529f\u80fd\uff1a\u968f\u673a\u5927\u5c0f\u3001\u957f\u5bbd\u6bd4\u88c1\u526a\u56fe\u7247<\/p>\n\n\n\n
              \"\"<\/div><\/figure>\n\n\n\n
                \n
              • size\uff1a\u6240\u9700\u88c1\u526a\u56fe\u7247\u5c3a\u5bf8<\/li>\n\n\n\n
              • scale\uff1a\u968f\u673a\u88c1\u526a\u9762\u79ef\u6bd4\u4f8b\uff0c\u9ed8\u8ba4\uff080.08\uff0c1\uff09<\/li>\n\n\n\n
              • ratio\uff1a\u968f\u673a\u957f\u5bbd\u6bd4\uff0c\u9ed8\u8ba4\uff083\/4\uff0c4\/3\uff09<\/li>\n\n\n\n
              • interpolation\uff1a\u63d2\u503c\u65b9\u6cd5<\/li>\n<\/ul>\n\n\n\n
                4. transforms.FiveCrop<\/h3>\n\n\n\n
                \u529f\u80fd\uff1a\u5728\u56fe\u50cf\u7684\u4e0a\u4e0b\u5de6\u53f3\u4ee5\u53ca\u4e2d\u5fc3\u88c1\u526a\u51fa\u5c3a\u5bf8\u4e3asize\u76845\u5f20\u56fe\u7247<\/p>\n\n\n\n
                \"\"<\/div><\/figure>\n\n\n\n
                  \n
                • size\uff1a\u6240\u9700\u88c1\u526a\u56fe\u7247\u5c3a\u5bf8<\/li>\n<\/ul>\n\n\n\n
                  5. transforms.TenCrop<\/h3>\n\n\n\n
                  \u529f\u80fd\uff1aTenCrop\u5bf9\u8fd95\u5f20\u56fe\u7247\u8fdb\u884c\u6c34\u5e73\u6216\u8005\u5782\u76f4\u955c\u50cf\u83b7\u5f9710\u5f20\u56fe\u7247<\/p>\n\n\n\n
                  \"\"<\/div><\/figure>\n\n\n\n
                    \n
                  • size\uff1a\u6240\u9700\u88c1\u526a\u56fe\u7247\u5c3a\u5bf8<\/li>\n\n\n\n
                  • vertical_flip\uff1a\u662f\u5426\u5782\u76f4\u7ffb\u8f6c<\/li>\n<\/ul>\n\n\n\n
                    \u7ffb\u8f6c Flip \u65cb\u8f6c Rotation<\/h2>\n\n\n\n
                    1. RandomHorizontalFlip<\/h3>\n\n\n\n
                    2. RandomVerticalFlip<\/h3>\n\n\n\n
                    \u529f\u80fd\uff1a\u4f9d\u6982\u7387\u6c34\u5e73\uff08\u5de6\u53f3\uff09\u6216\u5782\u76f4\uff08\u4e0a\u4e0b\uff09\u7ffb\u8f6c\u56fe\u7247<\/p>\n\n\n\n
                    \"\"<\/div><\/figure>\n\n\n\n
                      \n
                    • p\uff1a\u7ffb\u8f6c\u6982\u7387<\/li>\n<\/ul>\n\n\n\n
                      3. RandomRotation<\/h3>\n\n\n\n
                      \u529f\u80fd\uff1a\u968f\u673a\u65cb\u8f6c\u56fe\u7247<\/p>\n\n\n\n
                      \"\"<\/div><\/figure>\n\n\n\n
                        \n
                      • degrees\uff1a\u65cb\u8f6c\u89d2\u5ea6\n
                          \n
                        • \u5f53\u4e3aa\u65f6\uff0c\u5728\uff08-a\uff0ca\uff09\u4e4b\u95f4\u9009\u62e9\u65cb\u8f6c\u89d2\u5ea6<\/li>\n\n\n\n
                        • \u5f53\u4e3a\uff08a\uff0cb\uff09\u65f6\uff0c\u5728\uff08a\uff0cb\uff09\u4e4b\u95f4\u9009\u62e9\u65cb\u8f6c\u89d2\u5ea6<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n
                            \n
                          • resample\uff1a\u91cd\u91c7\u6837\u65b9\u6cd5<\/li>\n\n\n\n
                          • expand\uff1a\u662f\u5426\u6269\u5927\u56fe\u7247\uff0c\u4ee5\u4fdd\u6301\u539f\u56fe\u4fe1\u606f<\/li>\n\n\n\n
                          • center\uff1a\u65cb\u8f6c\u70b9\u8bbe\u7f6e\uff0c\u9ed8\u8ba4\u4e2d\u5fc3\u65cb\u8f6c<\/li>\n<\/ul>\n\n\n\n
                            \n\n\n\n
                            Transforms \u56fe\u50cf\u589e\u5f3a\uff08\u4e8c\uff09<\/h1>\n\n\n\n
                            \u56fe\u50cf\u53d8\u6362<\/h2>\n\n\n\n
                            1. Pad<\/h3>\n\n\n\n
                            \u529f\u80fd\uff1a\u5bf9\u56fe\u7247\u8fb9\u7f18\u8fdb\u884c\u586b\u5145<\/p>\n\n\n\n
                            \"\"<\/div><\/figure>\n\n\n\n
                              \n
                            • padding\uff1a\u8bbe\u7f6e\u586b\u5145\u5927\u5c0f\n
                                \n
                              • \u5f53\u4e3aa\u65f6\uff0c\u4e0a\u4e0b\u5de6\u53f3\u5747\u586b\u5145a\u4e2a\u50cf\u7d20<\/li>\n\n\n\n
                              • \u5f53\u4e3a\uff08a,b\uff09\u65f6\uff0c\u4e0a\u4e0b\u586b\u5145b\u4e2a\u50cf\u7d20\uff0c\u5de6\u53f3\u586b\u5145a\u4e2a\u50cf\u7d20<\/li>\n\n\n\n
                              • \u5f53\u4e3a\uff08a,b,c,d\uff09\u65f6\uff0c\u5de6\u3001\u4e0a\u3001\u53f3\u3001\u4e0b\u5206\u522b\u586b\u5145a,b,c,d<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n
                                  \n
                                • padding_mode\uff1a\u586b\u5145\u6a21\u5f0f\uff0c\u67094\u79cd\u6a21\u5f0f\uff0cconstant\u3001edge\u3001reflect\u548csymmetric<\/li>\n\n\n\n
                                • fill\uff1aconstant\u65f6\uff0c\u8bbe\u7f6e\u586b\u5145\u7684\u50cf\u7d20\u503c\uff0cRGB or Gray<\/li>\n<\/ul>\n\n\n\n
                                  2. ColorJitter<\/h3>\n\n\n\n
                                  \u529f\u80fd\uff1a\u8c03\u6574\u4eae\u5ea6\u3001\u5bf9\u6bd4\u5ea6\u3001\u9971\u548c\u5ea6\u548c\u8272\u76f8<\/p>\n\n\n\n
                                  \"\"<\/div><\/figure>\n\n\n\n
                                    \n
                                  • brightness\uff1a\u4eae\u5ea6\u8c03\u6574\u56e0\u5b50\n
                                      \n
                                    • \u5f53\u4e3aa\u65f6\uff0c\u4ece[max(0, 1-a), 1+a]\u4e2d\u968f\u673a\u9009\u62e9<\/li>\n\n\n\n
                                    • \u5f53\u4e3a\uff08a\uff0cb\uff09\u65f6\uff0c\u4ece[a\uff0cb]\u4e2d<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n
                                        \n
                                      • contrast\uff1a\u5bf9\u6bd4\u5ea6\u53c2\u6570\uff0c\u540cbrightness<\/li>\n\n\n\n
                                      • saturation\uff1a\u9971\u548c\u5ea6\u53c2\u6570\uff0c\u540cbrightness<\/li>\n\n\n\n
                                      • hue\uff1a\u8272\u76f8\u53c2\u6570\n
                                          \n
                                        • \u5f53\u4e3aa\u65f6\uff0c\u4ece[-a\uff0ca\uff3d\u4e2d\u9009\u62e9\u53c2\u6570\uff0c\u6ce8\uff1a0<=a<=0.5<\/li>\n\n\n\n
                                        • \u5f53\u4e3a\uff08a\uff0cb\uff09\u65f6\uff0c\u4ece[a\uff0cb]\u4e2d\u9009\u62e9\u53c2\u6570\uff0c\u6ce8\uff1a-0.5<=a<=b<=0.5<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n
                                          3. Grayscale<\/h3>\n\n\n\n
                                          4. RandomGrayscale<\/h3>\n\n\n\n
                                          \u529f\u80fd\uff1a\u4f9d\u6982\u7387\u5c06\u56fe\u7247\u8f6c\u6362\u4e3a\u7070\u5ea6\u56fe<\/p>\n\n\n\n
                                          \"\"<\/div><\/figure>\n\n\n\n
                                            \n
                                          • num_ouput_channels\uff1a\u8f93\u51fa\u901a\u9053\u6570\uff0c\u53ea\u80fd\u8bbe\u7f6e1\u62163<\/li>\n\n\n\n
                                          • p\uff1a\u6982\u7387\u503c\uff0c\u56fe\u50cf\u88ab\u8f6c\u6362\u4e3a\u7070\u5ea6\u56fe\u7684\u6982\u7387<\/li>\n<\/ul>\n\n\n\n
                                            5. RandomAffine<\/h3>\n\n\n\n
                                            \u529f\u80fd\uff1a\u5bf9\u56fe\u50cf\u8fdb\u884c\u4eff\u5c04\u53d8\u6362\uff0c\u4eff\u5c04\u53d8\u6362\u662f\u4e8c\u7ef4\u7684\u7ebf\u6027\u53d8\u6362\uff0c\u7531\u4e94\u79cd\u57fa\u672c\u539f\u5b50\u53d8\u6362\u6784\u6210\uff0c\u5206\u522b\u662f\u65cb\u8f6c\u3001\u5e73\u79fb\u3001\u7f29\u653e\u3001\u9519\u5207\u548c\u7ffb\u8f6c\u3002<\/p>\n\n\n\n
                                            \"\"<\/div><\/figure>\n\n\n\n
                                              \n
                                            • degrees\uff1a\u65cb\u8f6c\u89d2\u5ea6\u8bbe\u7f6e<\/li>\n\n\n\n
                                            • translate\uff1a\u5e73\u79fb\u533a\u95f4\u8bbe\u7f6e\uff0c\u5982\uff08a\uff0cb\uff09\uff0ca\u8bbe\u7f6e\u5bbd\uff08width\uff09\uff0cb\u8bbe\u7f6e\u9ad8\uff08height\uff09<\/li>\n\n\n\n
                                            • \u56fe\u50cf\u5728\u5bbd\u7ef4\u5ea6\u5e73\u79fb\u7684\u533a\u95f4\u4e3a -img_width *a< dx < img_width *a<\/li>\n<\/ul>\n\n\n\n
                                                \n
                                              • scale\uff1a\u7f29\u653e\u6bd4\u4f8b\uff08\u4ee5\u9762\u79ef\u4e3a\u5355\u4f4d\uff09<\/li>\n\n\n\n
                                              • fill_color\uff1a\u586b\u5145\u989c\u8272\u8bbe\u7f6e<\/li>\n\n\n\n
                                              • shear\uff1a\u9519\u5207\u89d2\u5ea6\u8bbe\u7f6e\uff0c\u6709\u6c34\u5e73\u9519\u5207\u548c\u5782\u76f4\u9519\u5207\n
                                                  \n
                                                • \u82e5\u4e3aa\uff0c\u5219\u4ec5\u5728x\u8f74\u9519\u5207\uff0c\u9519\u5207\u89d2\u5ea6\u5728\uff08-a\uff0ca\uff09\u4e4b\u95f4<\/li>\n\n\n\n
                                                • \u82e5\u4e3a\uff08a\uff0cb\uff09\uff0c\u5219a\u8bbe\u7f6ex\u8f74\u89d2\u5ea6\uff0cb\u8bbe\u7f6ey\u7684\u89d2\u5ea6<\/li>\n\n\n\n
                                                • \u82e5\u4e3a\uff08a\uff0cb\uff0cc\uff0cd\uff09\uff0c\u5219a\uff0cb\u8bbe\u7f6ex\u8f74\u89d2\u5ea6\uff0cc\uff0cd\u8bbe\u7f6ey\u8f74\u89d2\u5ea6<\/li>\n<\/ul>\n<\/li>\n\n\n\n
                                                • resample\uff1a\u91cd\u91c7\u6837\u65b9\u5f0f\uff0c\u6709NEAREST\uff0cBILINEAR\uff0cBICUBIC<\/li>\n<\/ul>\n\n\n\n
                                                  6. RandomErasing<\/h3>\n\n\n\n
                                                  \u529f\u80fd\uff1a\u5bf9\u56fe\u50cf\u8fdb\u884c\u968f\u673a\u906e\u6321<\/p>\n\n\n\n
                                                  \"\"<\/div><\/figure>\n\n\n\n
                                                    \n
                                                  • p\uff1a\u6982\u7387\u503c\uff0c\u6267\u884c\u8be5\u64cd\u4f5c\u7684\u6982\u7387<\/li>\n\n\n\n
                                                  • scale\uff1a\u906e\u6321\u533a\u57df\u7684\u9762\u79ef<\/li>\n\n\n\n
                                                  • ratio\uff1a\u906e\u6321\u533a\u57df\u957f\u5bbd\u6bd4<\/li>\n\n\n\n
                                                  • value\uff1a\u8bbe\u7f6e\u906e\u6321\u533a\u57df\u7684\u50cf\u7d20\u503c\uff0c\uff08R\uff0cG\uff0cB\uff09or\uff08Gray\uff09<\/li>\n<\/ul>\n\n\n\n
                                                    \"\"<\/div><\/figure>\n\n\n\n
                                                    7. transforms.Lambda<\/h3>\n\n\n\n
                                                    \u529f\u80fd\uff1a\u7528\u6237\u81ea\u5b9a\u4e49lambda\u65b9\u6cd5<\/p>\n\n\n\n
                                                    \"\"<\/div><\/figure>\n\n\n\n
                                                      \n
                                                    • lambd\uff1alambda\u533f\u540d\u51fd\u6570<\/li>\n\n\n\n
                                                    • lambda [arg1 [arg2, …, argn]] : expression<\/li>\n<\/ul>\n\n\n\n
                                                      \u5728\u4e4b\u524d\u6211\u4eec\u5c31\u7528\u8fc7\u8fd9\u4e2a\u51fd\u6570\uff1a<\/p>\n\n\n\n
                                                      \"\"<\/div><\/figure>\n\n\n\n
                                                      \n\n\n\n
                                                      transforms\u7684\u64cd\u4f5c<\/h2>\n\n\n\n
                                                      1. transforms.RandomChoice<\/h3>\n\n\n\n
                                                      \u529f\u80fd\uff1a\u4ece\u4e00\u7cfb\u5217transforms\u65b9\u6cd5\u4e2d\u968f\u673a\u6311\u9009\u4e00\u4e2a<\/p>\n\n\n\n
                                                      \"\"<\/div><\/figure>\n\n\n\n
                                                      2. transforms.RandomApply<\/h3>\n\n\n\n
                                                      \u529f\u80fd\uff1a\u4f9d\u636e\u6982\u7387\u6267\u884c\u4e00\u7ec4transforms\u64cd\u4f5c<\/p>\n\n\n\n
                                                      \"\"<\/div><\/figure>\n\n\n\n
                                                      3. transforms.RandomOrder<\/h3>\n\n\n\n
                                                      \u529f\u80fd\uff1a\u5bf9\u4e00\u7ec4transforms\u64cd\u4f5c\u6253\u4e71\u987a\u5e8f<\/p>\n\n\n\n
                                                      \"\"<\/div><\/figure>\n\n\n\n
                                                      \n\n\n\n
                                                      \u81ea\u5b9a\u4e49transforms<\/h2>\n\n\n\n
                                                      \"\"<\/div><\/figure>\n\n\n\n
                                                      \u81ea\u5b9a\u4e49transforms\u8981\u7d20\uff1a<\/p>\n\n\n\n
                                                        \n
                                                      1. \u4ec5\u63a5\u6536\u4e00\u4e2a\u53c2\u6570\uff0c\u8fd4\u56de\u4e00\u4e2a\u53c2\u6570<\/li>\n\n\n\n
                                                      2. \u6ce8\u610f\u4e0a\u4e0b\u6e38\u7684\u8f93\u51fa\u4e0e\u8f93\u5165<\/li>\n<\/ol>\n\n\n\n
                                                        \u901a\u8fc7\u7c7b\u5b9e\u73b0\u591a\u53c2\u6570\u4f20\u5165\uff1a<\/p>\n\n\n\n
                                                        \"\"<\/div><\/figure>\n\n\n\n
                                                        \n\n\n\n
                                                        \u793a\u4f8b\uff1a\u6912\u76d0\u566a\u58f0<\/p>\n\n\n\n
                                                        \u6912\u76d0\u566a\u58f0\u53c8\u79f0\u4e3a\u8109\u51b2\u566a\u58f0\uff0c\u662f\u4e00\u79cd\u968f\u673a\u51fa\u73b0\u7684\u767d\u70b9\u6216\u8005\u9ed1\u70b9\uff0c\u767d\u70b9\u79f0\u4e3a\u76d0\u566a\u58f0\uff0c\u9ed1\u8272\u4e3a\u6912\u566a\u58f0<\/p>\n\n\n\n
                                                        \u4fe1\u566a\u6bd4\uff08Signal-Noise Rate\uff0cSNR\uff09\u662f\u8861\u91cf\u566a\u58f0\u7684\u6bd4\u4f8b\uff0c\u56fe\u50cf\u4e2d\u4e3a\u56fe\u50cf\u50cf\u7d20\u7684\u5360\u6bd4<\/p>\n\n\n\n
                                                        \u4ece\u5de6\u5230\u53f3 \u4fe1\u566a\u6bd4\u8d8a\u6765\u8d8a\u5c0f\uff0c\u6912\u76d0\u566a\u58f0\u6bd4\u4f8b\u8d8a\u6765\u8d8a\u5927<\/p>\n\n\n\n
                                                        \"\"<\/div><\/figure>\n\n\n\n
                                                        \u81ea\u5b9a\u4e49\u4e00\u4e2a\u7c7b\uff1a<\/p>\n\n\n\n
                                                        \"\"<\/div><\/figure>\n\n\n\n
                                                        class AddPepperNoise(object):\n    \"\"\"\u589e\u52a0\u6912\u76d0\u566a\u58f0\n    Args:\n        snr \uff08float\uff09: Signal Noise Rate\n        p (float): \u6982\u7387\u503c\uff0c\u4f9d\u6982\u7387\u6267\u884c\u8be5\u64cd\u4f5c\n    \"\"\"\n\n    def __init__(self, snr, p=0.9):\n        assert isinstance(snr, float) or (isinstance(p, float))\n        self.snr = snr\n        self.p = p\n\n    def __call__(self, img):\n        \"\"\"\n        Args:\n            img (PIL Image): PIL Image\n        Returns:\n            PIL Image: PIL image.\n        \"\"\"\n        if random.uniform(0, 1) < self.p:\n            img_ = np.array(img).copy()\n            h, w, c = img_.shape\n            signal_pct = self.snr\n            noise_pct = (1 - self.snr)\n            mask = np.random.choice((0, 1, 2), size=(h, w, 1), p=[signal_pct, noise_pct\/2., noise_pct\/2.])\n            mask = np.repeat(mask, c, axis=2)\n            img_[mask == 1] = 255   # \u76d0\u566a\u58f0\n            img_[mask == 2] = 0     # \u6912\u566a\u58f0\n            return Image.fromarray(img_.astype('uint8')).convert('RGB')\n        else:\n            return img<\/code><\/pre>\n\n\n\n
                                                        \n\n\n\n
                                                        \u6570\u636e\u589e\u5f3a\u7b56\u7565<\/h2>\n\n\n\n
                                                        \"\"<\/div><\/figure>\n\n\n\n
                                                        \"\"<\/div><\/figure>\n\n\n\n
                                                        \u539f\u5219\uff1a\u8ba9\u8bad\u7ec3\u96c6\u4e0e\u6d4b\u8bd5\u96c6\u66f4\u63a5\u8fd1<\/p>\n\n\n\n
                                                          \n
                                                        • \u7a7a\u95f4\u4f4d\u7f6e\uff1a\u5e73\u79fb<\/li>\n\n\n\n
                                                        • \u8272\u5f69\uff1a\u7070\u5ea6\u56fe\uff0c\u8272\u5f69\u6296\u52a8<\/li>\n\n\n\n
                                                        • \u5f62\u72b6\uff1a\u4eff\u5c04\u53d8\u6362<\/li>\n\n\n\n
                                                        • \u4e0a\u4e0b\u6587\u573a\u666f\uff1a\u906e\u6321\uff0c\u586b\u5145<\/li>\n\n\n\n
                                                        • …..<\/li>\n<\/ul>\n\n\n\n
                                                          \n","protected":false},"excerpt":{"rendered":"
                                                          DataLoader\u4e0eDataSet \u4eba\u6c11\u5e01\u4e8c\u5206\u7c7b\u95ee\u9898 \u5047\u5982\u6211\u4eec\u8981\u5bf9100\u5143\u548c1\u5143\u4eba\u6c11\u5e01\u8fdb\u884c\u4e8c\u5206\u7c7b\uff0c\u90a3\u4e48\u6211\u4eec\u9700 […]<\/p>\n","protected":false},"author":1,"featured_media":98,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[10],"tags":[14,15],"class_list":["post-297","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-10","tag-pytorch","tag-15"],"_links":{"self":[{"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/posts\/297","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/comments?post=297"}],"version-history":[{"count":5,"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/posts\/297\/revisions"}],"predecessor-version":[{"id":352,"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/posts\/297\/revisions\/352"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/media\/98"}],"wp:attachment":[{"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/media?parent=297"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/categories?post=297"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.s7arry.top\/index.php\/wp-json\/wp\/v2\/tags?post=297"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}