In this tutorial, we'll see how the same API allows you to get a look at the inputs and outputs of your model, whether in the vision, text or tabular application. We'll go over a lot of different tasks and each time, grab some data in a
DataBunch with the data block API, see how to get a look at a few inputs with the
show_batch method, train an appropriate
Learner then use the
show_results method to see what the outputs of our model actually look like.
To quickly get access to all the vision functions inside fastai, we use the usual import statements.
from fastai.vision import *
Let's begin with our sample of the MNIST dataset.
mnist = untar_data(URLs.MNIST_TINY) tfms = get_transforms(do_flip=False)
It's set up with an imagenet structure so we use it to load our training and validation datasets, then label, transform, convert them into ImageDataBunch and finally, normalize them.
data = (ImageItemList.from_folder(mnist) .split_by_folder() .label_from_folder() .transform(tfms, size=32) .databunch() .normalize(imagenet_stats))
Once your data is properly set up in a
DataBunch, we can call
data.show_batch() to see what a sample of a batch looks like.
Note that the images were automatically de-normalized before being showed with their labels (inferred from the names of the folder). We can specify a number of rows if the default of 5 is too big, and we can also limit the size of the figure.
learn = create_cnn(data, models.resnet18, metrics=accuracy) learn.fit_one_cycle(1,1e-2) learn.save('mini_train')
Our model has quickly reache around 89% accuracy, now let's see its predictions on a sample of the validation set. For this, we use the
Since the validation set is usually sorted, we get only images belonging to the same class. We can then again specify a number of rows, a figure size, but also the dataset on which we want to make predictions.
learn.show_results(ds_type=DatasetType.Train, rows=4, figsize=(8,10))
Now let's try these on the planet dataset, which is a little bit different in the sense that each image can have multiple tags (and not jsut one label).
planet = untar_data(URLs.PLANET_TINY) planet_tfms = get_transforms(flip_vert=True, max_lighting=0.1, max_zoom=1.05, max_warp=0.)
Here each images is labelled in a file named 'labels.csv'. We have to add 'train' as a prefix to the filenames, '.jpg' as a suffix and he labels are separated by spaces.
data = (ImageItemList.from_csv(planet, 'labels.csv', folder='train', suffix='.jpg') .random_split_by_pct() .label_from_df(sep=' ') .transform(planet_tfms, size=128) .databunch() .normalize(imagenet_stats))
And we can have look at our data with
learn = create_cnn(data, models.resnet18) learn.fit_one_cycle(5,1e-2) learn.save('mini_train')
And to see actual predictions, we just have to run
For the next example, we are going to use the BIWI head pose dataset. On pictures of persons, we have to find the center of their face. For the fastai docs, we have built a small subsample of the dataset (200 images) and prepared a dictionary for the correspondance fielname to center.
biwi = untar_data(URLs.BIWI_SAMPLE) fn2ctr = pickle.load(open(biwi/'centers.pkl', 'rb'))
data = (ImageItemList.from_folder(biwi) .random_split_by_pct(seed=42) .label_from_func(lambda o:fn2ctr[o.name], label_cls=PointsItemList) .transform(get_transforms(), tfm_y=True, size=(120,160)) .databunch() .normalize(imagenet_stats))
data = (PointsItemList.from_folder(biwi) .random_split_by_pct(seed=42) .label_from_func(lambda o:fn2ctr[o.name]) .transform(get_transforms(), tfm_y=True, size=(120,160)) .databunch() .normalize(imagenet_stats))
Then we can have a first look at our data with