PJMixers-Images/Castollux-Long-Parquet

Castollux数据集包含使用Gemini API生成的图像及其对应的描述。描述主要是由模型`gemini-2.0-flash-thinking-exp-01-21`生成，少部分由`gemini-2.0-flash-thinking-exp-1219`生成。该数据集适用于图像到文本的任务，包含的图像可能具有较低的分辨率或压缩痕迹，这些痕迹有助于提高图像到文本的性能。但若用于文本到图像的训练，建议根据分辨率和压缩痕迹进行额外过滤。

The Castollux Dataset consists of images and their corresponding captions generated using the Gemini API. The captions are predominantly generated by the model `gemini-2.0-flash-thinking-exp-01-21`, with a smaller portion generated by `gemini-2.0-flash-thinking-exp-1219`. The dataset is intended for image-to-text tasks and includes images that may have lower resolution or compression artifacts, which are beneficial for improving image-to-text performance. However, for text-to-image training, additional filtering based on resolution and compression artifacts is recommended.