wehub-resource-sync
590 lines
26 KiB
Plaintext
590 lines
26 KiB
Plaintext
{
|
|
"cells": [
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"Deep Learning Models -- A collection of various deep learning architectures, models, and tips for TensorFlow and PyTorch in Jupyter Notebooks.\n",
|
|
"- Author: Sebastian Raschka\n",
|
|
"- GitHub Repository: https://github.com/rasbt/deeplearning-models"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 1,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Sebastian Raschka \n",
|
|
"\n",
|
|
"CPython 3.7.1\n",
|
|
"IPython 7.2.0\n",
|
|
"\n",
|
|
"torch 1.0.0\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"%load_ext watermark\n",
|
|
"%watermark -a 'Sebastian Raschka' -v -p torch"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"# Model Zoo -- Using PyTorch Dataset Loading Utilities for Custom Datasets (MNIST)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"This notebook provides an example for how to load an image dataset, stored as individual PNG files, using PyTorch's data loading utilities. For a more in-depth discussion, please see the official\n",
|
|
"\n",
|
|
"- [Data Loading and Processing Tutorial](http://pytorch.org/tutorials/beginner/data_loading_tutorial.html)\n",
|
|
"- [torch.utils.data](http://pytorch.org/docs/master/data.html) API documentation\n",
|
|
"\n",
|
|
"In this example, we are using the cropped version of the **Street View House Numbers (SVHN) Dataset**, which is available at http://ufldl.stanford.edu/housenumbers/. \n",
|
|
"\n",
|
|
"To execute the following examples, you need to download the 2 \".mat\" files \n",
|
|
"\n",
|
|
"- [train_32x32.mat](http://ufldl.stanford.edu/housenumbers/train_32x32.mat) (ca. 182 Mb, 73,257 images)\n",
|
|
"- [test_32x32.mat](http://ufldl.stanford.edu/housenumbers/test_32x32.mat) (ca. 65 Mb, 26,032 images)\n",
|
|
"\n",
|
|
"\n"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"## Imports"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 2,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"import pandas as pd\n",
|
|
"import numpy as np\n",
|
|
"import os\n",
|
|
"\n",
|
|
"import torch\n",
|
|
"from torch.utils.data import Dataset\n",
|
|
"from torch.utils.data import DataLoader\n",
|
|
"from torchvision import transforms\n",
|
|
"\n",
|
|
"import matplotlib.pyplot as plt\n",
|
|
"from PIL import Image\n",
|
|
"import scipy.io as sio\n",
|
|
"import imageio"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"## Dataset"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"The following function will convert the images from \".mat\" into individual \".png\" files. In addition, we will create CSV contained the image paths and associated class labels."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 3,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"def make_pngs(main_dir, mat_file, label):\n",
|
|
" \n",
|
|
" if not os.path.exists(main_dir):\n",
|
|
" os.mkdir(main_dir)\n",
|
|
" \n",
|
|
" sub_dir = os.path.join(main_dir, label)\n",
|
|
" if not os.path.exists(sub_dir):\n",
|
|
" os.mkdir(sub_dir)\n",
|
|
"\n",
|
|
" data = sio.loadmat(mat_file)\n",
|
|
"\n",
|
|
" X = np.transpose(data['X'], (3, 0, 1, 2))\n",
|
|
" y = data['y'].flatten()\n",
|
|
"\n",
|
|
" with open(os.path.join(main_dir, '%s_labels.csv' % label), 'w') as out_f:\n",
|
|
" for i, img in enumerate(X):\n",
|
|
" file_path = os.path.join(sub_dir, str(i) + '.png')\n",
|
|
" imageio.imwrite(os.path.join(file_path),\n",
|
|
" img)\n",
|
|
"\n",
|
|
" out_f.write(\"%d.png,%d\\n\" % (i, y[i]))\n",
|
|
"\n",
|
|
" \n",
|
|
"make_pngs(main_dir='svhn_cropped',\n",
|
|
" mat_file='train_32x32.mat',\n",
|
|
" label='train')\n",
|
|
" \n",
|
|
" \n",
|
|
"make_pngs(main_dir='svhn_cropped',\n",
|
|
" mat_file='test_32x32.mat',\n",
|
|
" label='test')"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 4,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"text/html": [
|
|
"<div>\n",
|
|
"<style scoped>\n",
|
|
" .dataframe tbody tr th:only-of-type {\n",
|
|
" vertical-align: middle;\n",
|
|
" }\n",
|
|
"\n",
|
|
" .dataframe tbody tr th {\n",
|
|
" vertical-align: top;\n",
|
|
" }\n",
|
|
"\n",
|
|
" .dataframe thead th {\n",
|
|
" text-align: right;\n",
|
|
" }\n",
|
|
"</style>\n",
|
|
"<table border=\"1\" class=\"dataframe\">\n",
|
|
" <thead>\n",
|
|
" <tr style=\"text-align: right;\">\n",
|
|
" <th></th>\n",
|
|
" <th>1</th>\n",
|
|
" </tr>\n",
|
|
" <tr>\n",
|
|
" <th>0</th>\n",
|
|
" <th></th>\n",
|
|
" </tr>\n",
|
|
" </thead>\n",
|
|
" <tbody>\n",
|
|
" <tr>\n",
|
|
" <th>0.png</th>\n",
|
|
" <td>1</td>\n",
|
|
" </tr>\n",
|
|
" <tr>\n",
|
|
" <th>1.png</th>\n",
|
|
" <td>9</td>\n",
|
|
" </tr>\n",
|
|
" <tr>\n",
|
|
" <th>2.png</th>\n",
|
|
" <td>2</td>\n",
|
|
" </tr>\n",
|
|
" <tr>\n",
|
|
" <th>3.png</th>\n",
|
|
" <td>3</td>\n",
|
|
" </tr>\n",
|
|
" <tr>\n",
|
|
" <th>4.png</th>\n",
|
|
" <td>2</td>\n",
|
|
" </tr>\n",
|
|
" </tbody>\n",
|
|
"</table>\n",
|
|
"</div>"
|
|
],
|
|
"text/plain": [
|
|
" 1\n",
|
|
"0 \n",
|
|
"0.png 1\n",
|
|
"1.png 9\n",
|
|
"2.png 2\n",
|
|
"3.png 3\n",
|
|
"4.png 2"
|
|
]
|
|
},
|
|
"execution_count": 4,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"df = pd.read_csv('svhn_cropped/train_labels.csv', header=None, index_col=0)\n",
|
|
"df.head()"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 5,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"text/html": [
|
|
"<div>\n",
|
|
"<style scoped>\n",
|
|
" .dataframe tbody tr th:only-of-type {\n",
|
|
" vertical-align: middle;\n",
|
|
" }\n",
|
|
"\n",
|
|
" .dataframe tbody tr th {\n",
|
|
" vertical-align: top;\n",
|
|
" }\n",
|
|
"\n",
|
|
" .dataframe thead th {\n",
|
|
" text-align: right;\n",
|
|
" }\n",
|
|
"</style>\n",
|
|
"<table border=\"1\" class=\"dataframe\">\n",
|
|
" <thead>\n",
|
|
" <tr style=\"text-align: right;\">\n",
|
|
" <th></th>\n",
|
|
" <th>1</th>\n",
|
|
" </tr>\n",
|
|
" <tr>\n",
|
|
" <th>0</th>\n",
|
|
" <th></th>\n",
|
|
" </tr>\n",
|
|
" </thead>\n",
|
|
" <tbody>\n",
|
|
" <tr>\n",
|
|
" <th>0.png</th>\n",
|
|
" <td>5</td>\n",
|
|
" </tr>\n",
|
|
" <tr>\n",
|
|
" <th>1.png</th>\n",
|
|
" <td>2</td>\n",
|
|
" </tr>\n",
|
|
" <tr>\n",
|
|
" <th>2.png</th>\n",
|
|
" <td>1</td>\n",
|
|
" </tr>\n",
|
|
" <tr>\n",
|
|
" <th>3.png</th>\n",
|
|
" <td>10</td>\n",
|
|
" </tr>\n",
|
|
" <tr>\n",
|
|
" <th>4.png</th>\n",
|
|
" <td>6</td>\n",
|
|
" </tr>\n",
|
|
" </tbody>\n",
|
|
"</table>\n",
|
|
"</div>"
|
|
],
|
|
"text/plain": [
|
|
" 1\n",
|
|
"0 \n",
|
|
"0.png 5\n",
|
|
"1.png 2\n",
|
|
"2.png 1\n",
|
|
"3.png 10\n",
|
|
"4.png 6"
|
|
]
|
|
},
|
|
"execution_count": 5,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"df = pd.read_csv('svhn_cropped/test_labels.csv', header=None, index_col=0)\n",
|
|
"df.head()"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"## Implementing a Custom Dataset Class"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"Now, we implement a custom `Dataset` for reading the images. The `__getitem__` method will\n",
|
|
"\n",
|
|
"1. read a single image from disk based on an `index` (more on batching later)\n",
|
|
"2. perform a custom image transformation (if a `transform` argument is provided in the `__init__` construtor)\n",
|
|
"3. return a single image and it's corresponding label"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 6,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"class SVHNDataset(Dataset):\n",
|
|
" \"\"\"Custom Dataset for loading cropped SVHN images\"\"\"\n",
|
|
" \n",
|
|
" def __init__(self, csv_path, img_dir, transform=None):\n",
|
|
" \n",
|
|
" df = pd.read_csv(csv_path, index_col=0, header=None)\n",
|
|
" self.img_dir = img_dir\n",
|
|
" self.csv_path = csv_path\n",
|
|
" self.img_names = df.index.values\n",
|
|
" self.y = df[1].values\n",
|
|
" self.transform = transform\n",
|
|
"\n",
|
|
" def __getitem__(self, index):\n",
|
|
" img = Image.open(os.path.join(self.img_dir,\n",
|
|
" self.img_names[index]))\n",
|
|
" \n",
|
|
" if self.transform is not None:\n",
|
|
" img = self.transform(img)\n",
|
|
" \n",
|
|
" label = self.y[index]\n",
|
|
" return img, label\n",
|
|
"\n",
|
|
" def __len__(self):\n",
|
|
" return self.y.shape[0]"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"Now that we have created our custom Dataset class, let us add some custom transformations via the `transforms` utilities from `torchvision`, we\n",
|
|
"\n",
|
|
"1. normalize the images (here: dividing by 255)\n",
|
|
"2. converting the image arrays into PyTorch tensors\n",
|
|
"\n",
|
|
"Then, we initialize a Dataset instance for the training images using the 'quickdraw_png_set1_train.csv' label file (we omit the test set, but the same concepts apply).\n",
|
|
"\n",
|
|
"Finally, we initialize a `DataLoader` that allows us to read from the dataset."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 7,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"# Note that transforms.ToTensor()\n",
|
|
"# already divides pixels by 255. internally\n",
|
|
"\n",
|
|
"custom_transform = transforms.Compose([#transforms.Grayscale(), \n",
|
|
" #transforms.Lambda(lambda x: x/255.),\n",
|
|
" transforms.ToTensor()])\n",
|
|
"\n",
|
|
"train_dataset = SVHNDataset(csv_path='svhn_cropped/train_labels.csv',\n",
|
|
" img_dir='svhn_cropped/train',\n",
|
|
" transform=custom_transform)\n",
|
|
"\n",
|
|
"test_dataset = SVHNDataset(csv_path='svhn_cropped/test_labels.csv',\n",
|
|
" img_dir='svhn_cropped/test',\n",
|
|
" transform=custom_transform)\n",
|
|
"\n",
|
|
"BATCH_SIZE=128\n",
|
|
"\n",
|
|
"\n",
|
|
"train_loader = DataLoader(dataset=train_dataset,\n",
|
|
" batch_size=BATCH_SIZE,\n",
|
|
" shuffle=True,\n",
|
|
" num_workers=4)\n",
|
|
"\n",
|
|
"test_loader = DataLoader(dataset=test_dataset,\n",
|
|
" batch_size=BATCH_SIZE,\n",
|
|
" shuffle=False,\n",
|
|
" num_workers=4)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"That's it, now we can iterate over an epoch using the train_loader as an iterator and use the features and labels from the training dataset for model training:"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"## Iterating Through the Custom Dataset"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 8,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Epoch: 1 | Batch index: 0 | Batch size: 128\n",
|
|
"Epoch: 2 | Batch index: 0 | Batch size: 128\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
|
|
"torch.manual_seed(0)\n",
|
|
"\n",
|
|
"num_epochs = 2\n",
|
|
"for epoch in range(num_epochs):\n",
|
|
"\n",
|
|
" for batch_idx, (x, y) in enumerate(train_loader):\n",
|
|
" \n",
|
|
" print('Epoch:', epoch+1, end='')\n",
|
|
" print(' | Batch index:', batch_idx, end='')\n",
|
|
" print(' | Batch size:', y.size()[0])\n",
|
|
" \n",
|
|
" x = x.to(device)\n",
|
|
" y = y.to(device)\n",
|
|
" break"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"Just to make sure that the batches are being loaded correctly, let's print out the dimensions of the last batch:"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 9,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"text/plain": [
|
|
"torch.Size([128, 3, 32, 32])"
|
|
]
|
|
},
|
|
"execution_count": 9,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"x.shape"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"As we can see, each batch consists of 128 images, just as specified. However, one thing to keep in mind though is that\n",
|
|
"PyTorch uses a different image layout (which is more efficient when working with CUDA); here, the image axes are \"num_images x channels x height x width\" (NCHW) instead of \"num_images height x width x channels\" (NHWC):"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"To visually check that the images that coming of the data loader are intact, let's swap the axes to NHWC and convert an image from a Torch Tensor to a NumPy array so that we can visualize the image via `imshow`:"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 10,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"text/plain": [
|
|
"torch.Size([32, 32, 3])"
|
|
]
|
|
},
|
|
"execution_count": 10,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"one_image = x[99].permute(1, 2, 0)\n",
|
|
"one_image.shape"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 11,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"image/png": "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\n",
|
|
"text/plain": [
|
|
"<Figure size 432x288 with 1 Axes>"
|
|
]
|
|
},
|
|
"metadata": {
|
|
"needs_background": "light"
|
|
},
|
|
"output_type": "display_data"
|
|
}
|
|
],
|
|
"source": [
|
|
"# note that imshow also works fine with scaled\n",
|
|
"# images in [0, 1] range.\n",
|
|
"plt.imshow(one_image.to(torch.device('cpu')));"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 12,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"pandas 0.23.4\n",
|
|
"scipy 1.1.0\n",
|
|
"matplotlib 3.0.2\n",
|
|
"imageio 2.4.1\n",
|
|
"torch 1.0.0\n",
|
|
"PIL.Image 5.3.0\n",
|
|
"numpy 1.15.4\n",
|
|
"torchvision 0.2.1\n",
|
|
"\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"%watermark -iv"
|
|
]
|
|
}
|
|
],
|
|
"metadata": {
|
|
"kernelspec": {
|
|
"display_name": "Python 3",
|
|
"language": "python",
|
|
"name": "python3"
|
|
},
|
|
"language_info": {
|
|
"codemirror_mode": {
|
|
"name": "ipython",
|
|
"version": 3
|
|
},
|
|
"file_extension": ".py",
|
|
"mimetype": "text/x-python",
|
|
"name": "python",
|
|
"nbconvert_exporter": "python",
|
|
"pygments_lexer": "ipython3",
|
|
"version": "3.7.1"
|
|
},
|
|
"toc": {
|
|
"nav_menu": {},
|
|
"number_sections": true,
|
|
"sideBar": true,
|
|
"skip_h1_title": false,
|
|
"title_cell": "Table of Contents",
|
|
"title_sidebar": "Contents",
|
|
"toc_cell": false,
|
|
"toc_position": {},
|
|
"toc_section_display": true,
|
|
"toc_window_display": false
|
|
}
|
|
},
|
|
"nbformat": 4,
|
|
"nbformat_minor": 2
|
|
}
|