from fastcore.all import *
from fastbook import search_images_ddg

def search_images(term, max_images=30):
    print(f"Searching for '{term}'")
    return search_images_ddg(term, max_images)

urls = search_images('bird photos', max_images=1)
urls[0]

Searching for 'bird photos'

'https://chilternchatter.com/wp-content/uploads/2018/01/RED-Bird.jpg'

Downloading image using the URL

from fastdownload import download_url
dest = 'bird.jpg'
download_url(urls[0], dest, show_progress=False)

from fastai.vision.all import *
im = Image.open(dest)
im.to_thumb(256,256)

png

# hide

download_url(search_images('forest photos', max_images=1)[0], 'forest.jpg', show_progress=False)
Image.open('forest.jpg').to_thumb(256,256)

Searching for 'forest photos'

png

searches = 'forest', 'bird'
path = Path('bird_or_not')
from time import sleep

for o in searches:
    dest = (path/o)
    dest.mkdir(exist_ok=True, parents=True)
    download_images(dest, urls=search_images(f'{o} photo'))
    sleep(2)
    download_images(dest, urls=search_images(f'{o} sun photo'))
    sleep(1)
    download_images(dest, urls=search_images(f'{o} shade photo'))
    sleep(2)
    resize_images(path/o, max_size=400, dest=path/o)

Searching for 'forest photo'
Searching for 'forest sun photo'
Searching for 'forest shade photo'
Searching for 'bird photo'
Searching for 'bird sun photo'
Searching for 'bird shade photo'

Step 2: Train our model

failed = verify_images(get_image_files(path))
failed.map(Path.unlink)
len(failed)

6

Using the DataLoaders

dls = DataBlock(
    blocks=(ImageBlock, CategoryBlock),
    get_items=get_image_files,
    splitter=RandomSplitter(valid_pct=0.2, seed=42),
    get_y=parent_label,
    item_tfms=[Resize(192, method='squish')]
).dataloaders(path, bs=32)

dls.show_batch(max_n=6)

png

learn = vision_learner(dls, resnet18, metrics=error_rate)
learn.fine_tune(3)

Downloading: "https://download.pytorch.org/models/resnet18-f37072fd.pth" to /home/kirubel/.cache/torch/hub/checkpoints/resnet18-f37072fd.pth
100%|██████████| 44.7M/44.7M [00:01<00:00, 44.1MB/s]
epoch train_loss valid_loss error_rate time
0 0.349523 0.017476 0.012903 00:42
epoch train_loss valid_loss error_rate time
0 0.055393 0.066558 0.012903 00:50
1 0.044839 0.027144 0.012903 00:52
2 0.026491 0.048613 0.012903 00:49

Step 3: Use our model (and build your own!)

is_bird,_,probs = learn.predict(PILImage.create('bird.jpg'))
print(f"This is a: {is_bird}.")
print(f"Probability it's a bird: {probs[0]:.4f}")

This is a: bird.
Probability it's a bird: 1.0000