File size: 1,802 Bytes
a81bb0b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
---
task_categories:
- image-classification
- unconditional-image-generation
size_categories:
- 10K<n<100K
---
# MNIST WebDataset PNG
The MNIST dataset with samples stored as PNG images and compiled into the WebDataset format.

## DALI/JAX Example
The following code shows how this dataset can be loaded into JAX arrays by DALI.
```python
from nvidia.dali import pipeline_def
import nvidia.dali.fn as fn
import nvidia.dali.types as types
from nvidia.dali.plugin.jax import DALIGenericIterator
from nvidia.dali.plugin.base_iterator import LastBatchPolicy

def get_data_iterator(batch_size, dataset_path):
    @pipeline_def(batch_size=batch_size, num_threads=4, device_id=0)
    def wds_pipeline():
        raw_image, ascii_label = fn.readers.webdataset(
            paths=dataset_path, 
            ext=['png', 'cls'], 
            missing_component_behavior='error',
        )
        image = fn.decoders.image(raw_image)
        ascii_shift = types.Constant(48).uint8()
        label = ascii_label - ascii_shift
        return image, label

    data_pipeline = wds_pipeline()
    data_iterator = DALIGenericIterator(
        pipelines=[data_pipeline], 
        output_map=['x', 'y'], 
        last_batch_policy=LastBatchPolicy.DROP
    )
    return data_iterator

data_iterator = get_data_iterator(
    batch_size=32, 
    dataset_path='data/mnist_webdataset_numpy_flat_9/data.tar'
)
batch = next(data_iterator)
x = batch['x']
y = batch['y']
print('x shape:', x.shape)
print('y shape:', y.shape)
print('y:', y[:, 0])
```

Output:
```
x shape: (32, 28, 28, 3)
y shape: (32, 1)
y: [5 0 4 1 9 2 1 3 1 4 3 5 3 6 1 7 2 8 6 9 4 0 9 1 1 2 4 3 2 7 3 8]
```

## Acknowledgements
- Yann LeCun, Courant Institute, NYU
- Corinna Cortes, Google Labs, New York
- Christopher J.C. Burges, Microsoft Research, Redmond