Platform For AI:Introduction to offline prediction

Read data from MaxCompute tables

The input table must include at least one column containing Base64-encoded binary data or URLs for the images. That column must be of the string type. Sample schema:

+------------------------------------------------------------------------------------+
| Field           | Type       | Label | Comment                                     |
+------------------------------------------------------------------------------------+
| id              | string     |       |                                             |
| url             | string     |       |                                             |
+------------------------------------------------------------------------------------+

Read data from OSS files

Each line in the OSS input file is either an OSS path or an HTTP URL pointing to an image:

oss://your/path/to/image.jpg
http://your.path/to/image.jpg

Custom input data

By default, reading from a MaxCompute table retrieves only the URL or Base64-encoded image data, and reading from OSS retrieves only the image URL. Both methods produce NumPy arrays in the {"image": np.ndarray} format — a single-input format that cannot meet all use cases.

The custom format can be the original OSS file format or JSON string format. Each line in the input file is a JSON string with one or more key-value pairs:

{"image":"http://your/path/to/image.jpg", "prior":"oss://your/path/to/prior.txt", "config": {"key1":1, "key2":"value2"}}
{"image":"http://your/path/to/image.jpg", "prior":"oss://your/path/to/prior.txt", "config": {"key2":1, "key2":"value2"}}

All key-value pairs are collected into a dictionary and passed to your custom predictor and process classes. The system automatically handles downloads and decoding:

• If a value is an OSS path or HTTP URL, the system downloads the file content using multiple threads and returns a file-like object. Call read() or readlines() to access the content.

• If the URL points to a file with an image extension, the system automatically decodes it and returns a numpy.ndarray.

The input is delivered to your predictor as an input_data dictionary:

input_dict = {
    "image": np.ndarray,          # Decoded image data
    "prior": file_like_object,    # Downloaded file as a file-like object
    "config": {"key1": 1, "key2": "value2"}  # Raw JSON sub-dictionary
}

Required parameters

Input and output

Compute and concurrency

Video parameters

Custom predictor and process

Reliability