remyxai/OpenSpaces_MC

OpenSpaces_MC数据集是一个使用VQASynth工具合成的空间视觉问答（VQA）数据集，它使用来自the cauldron数据集的前30K行本地化叙述的图片。这个数据集通过使用Molmo-7B-O模型代替Florence-2模型，提供了更优的图像字幕生成。与OpenSpaces数据集和相关数据集相比，OpenSpaces_MC在图像分布上具有更大的多样性，并且数据集中关于场景中物体相对位置的质量性距离估计查询多于定性查询。

The OpenSpaces_MC dataset is a spatial Visual Question Answering (VQA) dataset synthesized using VQASynth, with images from the first 30K rows of the localized narratives split of the cauldron dataset. This dataset provides improved image captioning using the Molmo-7B-O model instead of Florence-2. Compared to the OpenSpaces dataset and related datasets used to train SpaceLLaVA, OpenSpaces_MC offers greater diversity in image distribution. It also contains a higher number of qualitative distance estimation queries about the relative position of objects in a scene compared to qualitative queries.