| {"nbformat": 4, "cells": [{"source": "# Predict with pre-trained models\n\nThis tutorial explains how to recognize objects in an image with a pre-trained model, and how to perform feature extraction.\n\n## Prerequisites\n\nTo complete this tutorial, we need:\n\n- MXNet. See the instructions for your operating system in [Setup and Installation](http://mxnet.io/install/index.html)\n\n- [Matplotlib](https://matplotlib.org/) and [Jupyter Notebook](http://jupyter.org/index.html).\n\n```\n$ pip install matplotlib\n```\n\n## Loading\n\nWe first download a pre-trained ResNet 18 model that is trained on the ImageNet dataset with over 1 million images and one thousand classes. A pre-trained model contains two parts, a json file containing the model definition and a binary file containing the parameters. In addition, there may be a `synset.txt` text file for the labels.", "cell_type": "markdown", "metadata": {}}, {"source": "import mxnet as mx\npath='http://data.mxnet.io/models/imagenet/'\n[mx.test_utils.download(path+'resnet/18-layers/resnet-18-0000.params'),\n mx.test_utils.download(path+'resnet/18-layers/resnet-18-symbol.json'),\n mx.test_utils.download(path+'synset.txt')]", "cell_type": "code", "execution_count": null, "outputs": [], "metadata": {}}, {"source": "Next, we load the downloaded model. ", "cell_type": "markdown", "metadata": {}}, {"source": "# set the context on CPU, switch to GPU if there is one available\nctx = mx.cpu()", "cell_type": "code", "execution_count": null, "outputs": [], "metadata": {}}, {"source": "sym, arg_params, aux_params = mx.model.load_checkpoint('resnet-18', 0)\nmod = mx.mod.Module(symbol=sym, context=ctx, label_names=None)\nmod.bind(for_training=False, data_shapes=[('data', (1,3,224,224))], \n label_shapes=mod._label_shapes)\nmod.set_params(arg_params, aux_params, allow_missing=True)\nwith open('synset.txt', 'r') as f:\n labels = [l.rstrip() for l in f]", "cell_type": "code", "execution_count": null, "outputs": [], "metadata": {}}, {"source": "## Predicting\n\nWe first define helper functions for downloading an image and performing the\nprediction:", "cell_type": "markdown", "metadata": {}}, {"source": "%matplotlib inline\nimport matplotlib.pyplot as plt\nimport numpy as np\n# define a simple data batch\nfrom collections import namedtuple\nBatch = namedtuple('Batch', ['data'])\n\ndef get_image(url, show=False):\n # download and show the image. Remove query string from the file name.\n fname = mx.test_utils.download(url, fname=url.split('/')[-1].split('?')[0])\n img = mx.image.imread(fname)\n if img is None:\n return None\n if show:\n plt.imshow(img.asnumpy())\n plt.axis('off')\n # convert into format (batch, RGB, width, height)\n img = mx.image.imresize(img, 224, 224) # resize\n img = img.transpose((2, 0, 1)) # Channel first\n img = img.expand_dims(axis=0) # batchify\n return img\n\ndef predict(url):\n img = get_image(url, show=True)\n # compute the predict probabilities\n mod.forward(Batch([img]))\n prob = mod.get_outputs()[0].asnumpy()\n # print the top-5\n prob = np.squeeze(prob)\n a = np.argsort(prob)[::-1]\n for i in a[0:5]:\n print('probability=%f, class=%s' %(prob[i], labels[i]))", "cell_type": "code", "execution_count": null, "outputs": [], "metadata": {}}, {"source": "Now, we can perform prediction with any downloadable URL:", "cell_type": "markdown", "metadata": {}}, {"source": "predict('https://github.com/dmlc/web-data/blob/master/mxnet/doc/tutorials/python/predict_image/cat.jpg?raw=true')", "cell_type": "code", "execution_count": null, "outputs": [], "metadata": {}}, {"source": "predict('https://github.com/dmlc/web-data/blob/master/mxnet/doc/tutorials/python/predict_image/dog.jpg?raw=true')", "cell_type": "code", "execution_count": null, "outputs": [], "metadata": {}}, {"source": "## Feature extraction\n\nBy feature extraction, we mean presenting the input images by the output of an internal layer rather than the last softmax layer. These outputs, which can be viewed as the feature of the raw input image, can then be used by other applications such as object detection.\n\nWe can use the ``get_internals`` method to get all internal layers from a Symbol.", "cell_type": "markdown", "metadata": {}}, {"source": "# list the last 10 layers\nall_layers = sym.get_internals()\nall_layers.list_outputs()[-10:]", "cell_type": "code", "execution_count": null, "outputs": [], "metadata": {}}, {"source": "```\n['bn1_moving_var',\n 'bn1_output',\n 'relu1_output',\n 'pool1_output',\n 'flatten0_output',\n 'fc1_weight',\n 'fc1_bias',\n 'fc1_output',\n 'softmax_label',\n 'softmax_output']\n ```\n\nAn often used layer for feature extraction is the one before the last fully connected layer. For ResNet, and also Inception, it is the flattened layer with name `flatten0` which reshapes the 4-D convolutional layer output into 2-D for the fully connected layer. The following source code extracts a new Symbol which outputs the flattened layer and creates a model.", "cell_type": "markdown", "metadata": {}}, {"source": "fe_sym = all_layers['flatten0_output']\nfe_mod = mx.mod.Module(symbol=fe_sym, context=ctx, label_names=None)\nfe_mod.bind(for_training=False, data_shapes=[('data', (1,3,224,224))])\nfe_mod.set_params(arg_params, aux_params)", "cell_type": "code", "execution_count": null, "outputs": [], "metadata": {}}, {"source": "We can now invoke `forward` to obtain the features:", "cell_type": "markdown", "metadata": {}}, {"source": "img = get_image('https://github.com/dmlc/web-data/blob/master/mxnet/doc/tutorials/python/predict_image/cat.jpg?raw=true')\nfe_mod.forward(Batch([img]))\nfeatures = fe_mod.get_outputs()[0]\nprint('Shape',features.shape)\nprint(features.asnumpy())\nassert features.shape == (1, 512)", "cell_type": "code", "execution_count": null, "outputs": [], "metadata": {}}, {"source": "\n<!-- INSERT SOURCE DOWNLOAD BUTTONS -->\n\n", "cell_type": "markdown", "metadata": {}}], "metadata": {"display_name": "", "name": "", "language": "python"}, "nbformat_minor": 2} |