Add files via upload

Beating University of Oxford-IIIT Pet Model.ipynb

@@ -0,0 +1 @@
+{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"name":"Beating University of Oxford-IIIT Pet Model.ipynb","provenance":[],"collapsed_sections":["-zPn180gQ0sz","wmuF2z7vV15M"]},"kernelspec":{"name":"python3","display_name":"Python 3"}},"cells":[{"cell_type":"code","metadata":{"id":"SYOWmy55OMKh","colab_type":"code","colab":{}},"source":[""],"execution_count":0,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"J7eUhwn-Ppx4","colab_type":"text"},"source":[""]},{"cell_type":"markdown","metadata":{"id":"-zPn180gQ0sz","colab_type":"text"},"source":["# University of Oxford's IIIT Pet Model\n","The Oxford-IIIT Pet dataset was created for the fine-grained categorisation problem of identifying the family and breed of pets (both;cats and dogs). Three different tasks and corresponding baseline algorithms were proposed and investigated, obtaining a mere 42% probability.\n","[Comparison of 3 models of Oxford's IIT pet ](https://http://www.robots.ox.ac.uk/~vgg/publications/2012/parkhi12a/parkhi12a.pdf)"]},{"cell_type":"markdown","metadata":{"id":"sqoPsSSVPo_Y","colab_type":"text"},"source":[""]},{"cell_type":"markdown","metadata":{"id":"PLpJAkVTTeze","colab_type":"text"},"source":["# Using CNN to build an Image- Classifier Model"]},{"cell_type":"code","metadata":{"id":"T7R6ghvhTyMp","colab_type":"code","colab":{}},"source":["# Going to start with the following three lines; ensuring that any edits to libraries one make get reloaded here automatically, and also that any charts or images displayed are shown in this notebook.\n","\n","%reload_ext autoreload\n","%autoreload 2\n","%matplotlib inline\n","\n","# importing the necessary packages\n","\n","from fastai.vision import *\n","from fastai.metrics import error_rate"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"mNYGRcsPUc93","colab_type":"code","colab":{}},"source":["bs = 64\n","# bs = 16   # uncomment this line if you run out of memory even after clicking Kernel->Restart"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"jCkGDhQAUhsA","colab_type":"code","colab":{}},"source":["# going to use the untar_data function to which we must pass a URL as an argument and which will download and extract the data.\n","\n","help(untar_data)"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"7zEiasq1Uuie","colab_type":"code","colab":{}},"source":["path = untar_data(URLs.PETS); path"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"EfJ7wG42Uvl5","colab_type":"code","colab":{}},"source":["path.ls()"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"5OyK8SbGUxvC","colab_type":"code","colab":{}},"source":["fnames = get_image_files(path_img)\n","fnames[:5]"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"Or1eM3YyU2_Y","colab_type":"code","colab":{}},"source":["np.random.seed(2)\n","pat = r'/([^/]+)_\\d+.jpg$'"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"whyFYrkpU6_Y","colab_type":"code","colab":{}},"source":["data = ImageDataBunch.from_name_re(path_img, fnames, pat, ds_tfms=get_transforms(), size=224, bs=bs\n","                                  ).normalize(imagenet_stats)"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"UN_DUI1vU-Pe","colab_type":"code","colab":{}},"source":["data.show_batch(rows=3, figsize=(7,6))"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"fOaCkJ01VA6m","colab_type":"code","colab":{}},"source":["print(data.classes)\n","len(data.classes),data.c"],"execution_count":0,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"j6iwOBccVYWe","colab_type":"text"},"source":["## Using restnet34"]},{"cell_type":"code","metadata":{"id":"OwfvFsd5Vflc","colab_type":"code","colab":{}},"source":["learn = cnn_learner(data, models.resnet34, metrics=error_rate)"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"X_G94sexVkLb","colab_type":"code","colab":{}},"source":["learn.model"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"8CLr8Y6aVmab","colab_type":"code","colab":{}},"source":["learn.fit_one_cycle(4)"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"Iy722NDSVpDs","colab_type":"code","colab":{}},"source":["learn.save('stage-1')"],"execution_count":0,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"wmuF2z7vV15M","colab_type":"text"},"source":["### Interpretation"]},{"cell_type":"code","metadata":{"id":"yTdWiBH-V7Fs","colab_type":"code","colab":{}},"source":["interp = ClassificationInterpretation.from_learner(learn)\n","\n","losses,idxs = interp.top_losses()\n","\n","len(data.valid_ds)==len(losses)==len(idxs)"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"fbMFI0uDV9-n","colab_type":"code","colab":{}},"source":["interp.plot_top_losses(9, figsize=(15,11))"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"TilWU74QWAcX","colab_type":"code","colab":{}},"source":["interp.plot_confusion_matrix(figsize=(12,12), dpi=60)"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"LvsSvLI2WmRE","colab_type":"code","colab":{}},"source":["interp.most_confused(min_val=2)"],"execution_count":0,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"30GKIKP8Wpim","colab_type":"text"},"source":["### Unfreezing and fine-tuning"]},{"cell_type":"code","metadata":{"id":"WcCSQ7fdWyU9","colab_type":"code","colab":{}},"source":["learn.unfreeze()"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"LHPmlEoFW1dY","colab_type":"code","colab":{}},"source":["learn.fit_one_cycle(1)"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"PzBg_c0RW35c","colab_type":"code","colab":{}},"source":["learn.load('stage-1');"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"MjA6ZE2XW6SD","colab_type":"code","colab":{}},"source":["learn.lr_find()"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"zQp7g6APW8Ry","colab_type":"code","colab":{}},"source":["learn.recorder.plot()"],"execution_count":0,"outputs":[]},{"cell_type":"code","metadata":{"id":"y4UqhyDxW-nd","colab_type":"code","colab":{}},"source":["learn.unfreeze()\n","learn.fit_one_cycle(2, max_lr=slice(1e-6,1e-4))"],"execution_count":0,"outputs":[]},{"cell_type":"markdown","metadata":{"id":"xm8NQAdPXdrO","colab_type":"text"},"source":["# CNN- a fine development"]},{"cell_type":"markdown","metadata":{"id":"A_wT44RHXxol","colab_type":"text"},"source":["*A regularized version of multilayer perceptrons*, convolutional Neural Network i a class of deep neural networks, most commonly applied to analyzing visual imagery.\n","Also known as **shift invariant or space invariant artificial neural networks (SIANN)**, based on their shared-weights architecture and translation invariance characteristics, Convolutional networks were inspired by biological processes in that the connectivity pattern between neurons resembles the organization of the animal visual cortex.\n","**CNNs use relatively little pre-processing compared to other image classification algorithms implying that the network learns the filters that in traditional algorithms were hand-engineered. This independence from prior knowledge and human effort in feature design is a major advantage.**"]}]}