loganbolton/sketchvlm-physics-ball-drop
收藏Hugging Face2026-04-29 更新2026-05-03 收录
下载链接：
https://hf-mirror.com/datasets/loganbolton/sketchvlm-physics-ball-drop
下载链接
链接失效反馈官方服务：
资源简介：
SketchVLM是一个无需训练、模型无关的框架，使视觉语言模型(VLMs)能够在输入图像上生成非破坏性、可编辑的SVG覆盖层，以直观解释其推理过程。Physics Ball Drop数据集(基于PHYRE)是为评估视觉推理而开发的基准测试之一。在此任务中，模型必须预测球通过各种障碍物的轨迹，并提供路径的视觉解释。
SketchVLM is a training-free, model-agnostic framework that enables vision-language models (VLMs) to produce non-destructive, editable SVG overlays on input images to visually explain their reasoning. The Physics Ball Drop dataset (based on PHYRE) is one of the benchmarks developed to evaluate visual reasoning. In this task, models must predict the trajectory of a ball through various obstacles and provide a visual explanation of the path.
提供机构：
loganbolton
