TESS-Computer/tess-atari-15hz-384

TESS-Atari Stage 1 - Preprocessed (15Hz, 384x384)是一个用于训练视觉语言动作模型（VLA）的预处理数据集。该数据集包含1,340,293个样本，图像已预先调整为384x384分辨率，动作速率为15Hz。数据集的特征包括二进制格式的PNG图像、Lumine风格的分块动作令牌、游戏名称、人类玩家试验编号、试验中的帧索引和固定为384的图像大小。数据集的目的是通过预先调整图像大小来加速训练过程，避免在训练过程中实时调整大小的CPU瓶颈。

TESS-Atari Stage 1 - Preprocessed (15Hz, 384x384) is a preprocessed dataset for training vision-language-action models (VLA). The dataset contains 1,340,293 samples with images pre-resized to 384x384 resolution and an action rate of 15Hz. The features of the dataset include image_bytes (PNG images in binary format), action (Lumine-style chunked action tokens), game (game name), trial_id (human player trial number), frame_idx (frame index in trial), and image_size (always 384). The purpose of the dataset is to speed up the training process by pre-resizing the images, avoiding the CPU bottleneck of on-the-fly resizing during training.