Skip to content

Read as MULLER object

1. Loading Multimodal Files into MULLER Objects

MULLER provides the muller.read() interface for loading files from different storage backends via the path argument.
Below, we use image loading as an example.

Example 1: Loading from a Local File Path 

>>> img_1 = muller.read(path="/your/data/path/xxx.jpg")
>>> img_1.array  # Access the tensor representation of the multimodal object via `.array`.
array([[[183, 160, 118],
        [223, 201, 160],
        [211, 189, 150],
        ...,
        [143, 146,  79],
        [186, 189, 122],
        [142, 145,  78]]], shape=(400, 500, 3), dtype=uint8)

>>> img_1.shape  # Retrieve the tensor shape via the `.shape` attribute.
(400, 500, 3)  # This JPEG image has a resolution of 400×500 (height × width) with three RGB channels.

>>> img_1.dtype  # Retrieve the underlying data type via the `.dtype` attribute.
'uint8'  # JPEG images represent pixel values in the range [0, 255], and are therefore stored as uint8.

Example 2: Loading from an HTTP/HTTPS URL
(If access is restricted, the creds argument can be used to provide proxy settings.) 

>>> credential_2 = {
...     "proxies": {
...         "http": "http://account:pwd@proxy.xxx.com:port",
...         "https": "http://account:pwd@proxy.xxx.com:port"
...     }
... }
>>> img_2 = muller.read(
...     path="https://xxx.com/xxx.jpg",
...     creds=credential_2
... )

Example 3: Loading from a ROMA/S3 object storage bucket 

>>> credential_3 = {
...     "bucket_name": "bucket-name",
...     "region": "xxx",
...     "app_token": "xxx",
...     "vendor": "xxx"
... }
>>> img_3 = muller.read(
...     path="roma://your/data/path/xxx.jpg",
...     creds=credential_3
... )

  • muller.read() currently supports multiple multimodal file types, including images, audio, and video, covering more than 20 file formats.
  • The optional creds argument can be used to access remote or cloud storage backends (e.g., HTTP/HTTPS, ROMA, OBS), and may include authentication or connection information such as proxies, bucket_name, access_key, secret_key, and token.
  • For the complete list of input parameters, supported file types, and usage details of muller.read(), please refer to the API documentation.