nouveau 1.0.1

public domain art nouveau image data: lazy-downloading datasets of Morris, Mucha, and Driscoll

Art Nouveau Design Data

Public domain lazy-downloading data samples.

• Mucha
• Morris
• Driscoll — Tiffany Studios lamp and glass designs (CC BY-SA 4.0)

Example Use

import nouveau

# random image from all datasets
img = next(nouveau)
>> <PIL Image>

# metadata for the last image returned by next()
nouveau.details
>> {'year': 1897, 'name': 'Zodiac', 'filename': '1897-Zodiac.jpg'}

Indexed access — randomly assigned on first read, cached thereafter

nouveau[0]
>> <PIL Image>
nouveau[0]  # same image
>> <PIL Image>

Sample a finite batch

import itertools
batch = list(itertools.islice(nouveau, 8))

Infinite stream

for img in nouveau:
    process(img)

Collection Access & Sampling

morris   = nouveau.collections.Morris()
mucha    = nouveau.collections.Mucha()
driscoll = nouveau.collections.Driscoll()

# pandas index of metadata
morris.index.head()
>> shows pandas dataframe head with year, name, filename columns

# random image from a specific collection
driscoll.sample()
>> <PIL Image>

# iterate as PIL Images (aligned with index by position)
for i, img in enumerate(morris):
    print(morris.index.iloc[i]['name'], img.size)

# access a single item by position (returns dict with numpy image)
morris[0]
>> {'year': 1862,
>>  'name': 'Fruit-Blue',
>>  'filename': '1862-Fruit-Blue.jpg',
>>  'image': array([[[254, 253, 249], ...

# display by index or name
morris.show(1)
>> <pyplot image>

morris.show('Fruit-Blue')
>> <pyplot image>

# get a PIL Image by index or name
morris.toPIL(0)
>> <PIL Image>

# PyTorch dataset (returns [tensor, name, year] per item)
tensors = morris.to_torch()
tensors[0][0].shape
>> torch.Size([3, 1987, 1586])

---

license: cc-by-sa-4.0 task_categories:

• image-classification pretty_name: Art Nouveau Design Data size_categories:
• n<1K tags:
• morris
• mucha
• driscoll
• tiffany

---

Licenses

Collection	License
Morris	Unlicense
Mucha	Unlicense
Driscoll	CC BY-SA 4.0

The package code is Unlicense. Image data licenses vary by collection as noted above. The overall dataset is distributed under CC BY-SA 4.0 as the most restrictive license present.