{
  "id": 119332,
  "title": "dataloader",
  "url": "/competitions/understanding_cloud_organization/discussion/119332",
  "author_name": "byr_syx",
  "post_date": "2019-11-28T04:11:37.056000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>def default_loader(id, root, df, shape):\n    image_name = os.path.join(root,'train_images','{}.jpg').format(id)\n    img = cv2.imread(image_name)\n    img = cv2.resize(img, shape)   #(512,1024) \n    mask = make_mask(df, image_name, shape) #1400,2100,4\n    mask = cv2.resize(mask, shape)</p>\n\n<h1>augmented,transform</h1>\n\n<pre><code>img = randomHueSaturationValue(img,\n                               hue_shift_limit=(-35, 35),\n                               sat_shift_limit=(-20, 20),\n                               val_shift_limit=(-20, 20))\n\nimg, mask = randomShiftScaleRotate(img, mask,\n                                   shift_limit=(-0.1, 0.1),\n                                   scale_limit=(-0.1, 0.1),\n                                   aspect_limit=(-0.1, 0.1),\n                                   rotate_limit=(-0, 0))\nimg, mask = randomHorizontalFlip(img, mask)\nimg, mask = randomVerticleFlip(img, mask)\n#img, mask = randomRotate90(img, mask) # w != h\n</code></pre>\n\n<p>#preprocessing <br>\n    img = np.array(img, np.float32).transpose(2,0,1)/255.0 * 3.2 - 1.6\n    mask = np.array(mask, np.float32).transpose(2,0,1)</p>\n\n<h1>ToTensor in the <strong>getitem</strong>()</h1>\n\n<pre><code>return img, mask\n</code></pre>\n\n<p>class ImageFolder(data.Dataset):\n    def <strong>init</strong>(self, trainlist, root, df, shape):\n        self.ids = trainlist\n        self.loader = default_loader\n        self.root = root\n        self.df = df\n        self.shape = shape\n    def <strong>getitem</strong>(self, index):\n        id = self.ids[index]\n        img, mask = self.loader(id, self.root, self.df, self.shape)\n        #ToTensor\n        img = torch.Tensor(img)\n        mask = torch.Tensor(mask)\n        #print 'size(img)', img.size()\n        #print 'size(mask)', mask.size()\n        return img, mask\n    def <strong>len</strong>(self):\n        return len(self.ids)</p>",
  "messages": [
    {
      "id": 683079,
      "postDate": "2019-11-28T04:11:37.057Z",
      "content": "<p>def default_loader(id, root, df, shape):\n    image_name = os.path.join(root,'train_images','{}.jpg').format(id)\n    img = cv2.imread(image_name)\n    img = cv2.resize(img, shape)   #(512,1024) \n    mask = make_mask(df, image_name, shape) #1400,2100,4\n    mask = cv2.resize(mask, shape)</p>\n\n<h1>augmented,transform</h1>\n\n<pre><code>img = randomHueSaturationValue(img,\n                               hue_shift_limit=(-35, 35),\n                               sat_shift_limit=(-20, 20),\n                               val_shift_limit=(-20, 20))\n\nimg, mask = randomShiftScaleRotate(img, mask,\n                                   shift_limit=(-0.1, 0.1),\n                                   scale_limit=(-0.1, 0.1),\n                                   aspect_limit=(-0.1, 0.1),\n                                   rotate_limit=(-0, 0))\nimg, mask = randomHorizontalFlip(img, mask)\nimg, mask = randomVerticleFlip(img, mask)\n#img, mask = randomRotate90(img, mask) # w != h\n</code></pre>\n\n<p>#preprocessing <br>\n    img = np.array(img, np.float32).transpose(2,0,1)/255.0 * 3.2 - 1.6\n    mask = np.array(mask, np.float32).transpose(2,0,1)</p>\n\n<h1>ToTensor in the <strong>getitem</strong>()</h1>\n\n<pre><code>return img, mask\n</code></pre>\n\n<p>class ImageFolder(data.Dataset):\n    def <strong>init</strong>(self, trainlist, root, df, shape):\n        self.ids = trainlist\n        self.loader = default_loader\n        self.root = root\n        self.df = df\n        self.shape = shape\n    def <strong>getitem</strong>(self, index):\n        id = self.ids[index]\n        img, mask = self.loader(id, self.root, self.df, self.shape)\n        #ToTensor\n        img = torch.Tensor(img)\n        mask = torch.Tensor(mask)\n        #print 'size(img)', img.size()\n        #print 'size(mask)', mask.size()\n        return img, mask\n    def <strong>len</strong>(self):\n        return len(self.ids)</p>",
      "rawMarkdown": "def default_loader(id, root, df, shape):\n    image_name = os.path.join(root,'train_images','{}.jpg').format(id)\n    img = cv2.imread(image_name)\n    img = cv2.resize(img, shape)   #(512,1024) \n    mask = make_mask(df, image_name, shape) #1400,2100,4\n    mask = cv2.resize(mask, shape)\n#augmented,transform\n    img = randomHueSaturationValue(img,\n                                   hue_shift_limit=(-35, 35),\n                                   sat_shift_limit=(-20, 20),\n                                   val_shift_limit=(-20, 20))\n\n    img, mask = randomShiftScaleRotate(img, mask,\n                                       shift_limit=(-0.1, 0.1),\n                                       scale_limit=(-0.1, 0.1),\n                                       aspect_limit=(-0.1, 0.1),\n                                       rotate_limit=(-0, 0))\n    img, mask = randomHorizontalFlip(img, mask)\n    img, mask = randomVerticleFlip(img, mask)\n    #img, mask = randomRotate90(img, mask) # w != h\n #preprocessing   \n    img = np.array(img, np.float32).transpose(2,0,1)/255.0 * 3.2 - 1.6\n    mask = np.array(mask, np.float32).transpose(2,0,1)\n#ToTensor in the __getitem__()\n    return img, mask\n\nclass ImageFolder(data.Dataset):\n    def __init__(self, trainlist, root, df, shape):\n        self.ids = trainlist\n        self.loader = default_loader\n        self.root = root\n        self.df = df\n        self.shape = shape\n    def __getitem__(self, index):\n        id = self.ids[index]\n        img, mask = self.loader(id, self.root, self.df, self.shape)\n        #ToTensor\n        img = torch.Tensor(img)\n        mask = torch.Tensor(mask)\n        #print 'size(img)', img.size()\n        #print 'size(mask)', mask.size()\n        return img, mask\n    def __len__(self):\n        return len(self.ids)"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "683079": "def default_loader(id, root, df, shape):\n    image_name = os.path.join(root,'train_images','{}.jpg').format(id)\n    img = cv2.imread(image_name)\n    img = cv2.resize(img, shape)   #(512,1024) \n    mask = make_mask(df, image_name, shape) #1400,2100,4\n    mask = cv2.resize(mask, shape)\n#augmented,transform\n    img = randomHueSaturationValue(img,\n                                   hue_shift_limit=(-35, 35),\n                                   sat_shift_limit=(-20, 20),\n                                   val_shift_limit=(-20, 20))\n\n    img, mask = randomShiftScaleRotate(img, mask,\n                                       shift_limit=(-0.1, 0.1),\n                                       scale_limit=(-0.1, 0.1),\n                                       aspect_limit=(-0.1, 0.1),\n                                       rotate_limit=(-0, 0))\n    img, mask = randomHorizontalFlip(img, mask)\n    img, mask = randomVerticleFlip(img, mask)\n    #img, mask = randomRotate90(img, mask) # w != h\n #preprocessing   \n    img = np.array(img, np.float32).transpose(2,0,1)/255.0 * 3.2 - 1.6\n    mask = np.array(mask, np.float32).transpose(2,0,1)\n#ToTensor in the __getitem__()\n    return img, mask\n\nclass ImageFolder(data.Dataset):\n    def __init__(self, trainlist, root, df, shape):\n        self.ids = trainlist\n        self.loader = default_loader\n        self.root = root\n        self.df = df\n        self.shape = shape\n    def __getitem__(self, index):\n        id = self.ids[index]\n        img, mask = self.loader(id, self.root, self.df, self.shape)\n        #ToTensor\n        img = torch.Tensor(img)\n        mask = torch.Tensor(mask)\n        #print 'size(img)', img.size()\n        #print 'size(mask)', mask.size()\n        return img, mask\n    def __len__(self):\n        return len(self.ids)"
  }
}