# Drawing with LLMs — Data

来源：https://www.kaggle.com/competitions/drawing-with-llms/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

The competition test data comprises about 500 descriptions of everyday objects and scenes from a variety of domains. Your goal is to create a model that generates [SVG](https://www.w3.org/TR/SVG2/Overview.html) depictions of these descriptions.

The descriptions have the following properties:
- The descriptions are of common, generic subjects. No brand name or trademark or personal name occurs in any description. No people, even in generic form, occur in any description.
- The subjects described span about a dozen categories. Three of these categories, `landscapes`, `abstract`, and `fashion`, are present in each of the training, public test, and private test sets. The remaining categories in the public and private sets are unique to each set. More than half of the descriptions come from the three shared categories.
- No description has more than 200 characters. The average length is around 50 characters.

## Files

*   **train.csv** - A set of representative descriptions within the `landscapes`, `abstract`, and `fashion` categories.
*   **kaggle_evaluation/** - A Python package with competition helpers, such as the `test(Model)` function that simulates the evaluation system using a mock `test.csv`.
*   **kaggle_evaluation/test.csv** - An example set of test data to help you author your submissions. When your submission is scored, this placeholder test data will be replaced with the actual test data.

## Evaluation

The dataset contains a Python package in `kaggle_evaluation/` with competition helpers such as a `test(Model)` function. The `test.csv` file is used by this Python package in Kaggle’s scoring system. Kaggle’s scoring model reads this file and uses it as inputs to calling your  `predict()` function for generating submissions on the evaluation set.