Datasetpublic

loganbolton/sketchvlm-connect-dots

SketchVLM: Physics Ball Drop Dataset This dataset is part of the SketchVLM framework, which enables Vision-Language Models (VLMs) to produce non-destructive, editable SVG overlays on input images to visually explain their reasoning. The Physics Ball Drop benchmark specifically evaluates a model's ability to predict the trajectory of a ball through obstacles (based on PHYRE). Paper: SketchVLM: Vision language models can annotate images to explain thoughts and guide users Project… See the full description on the dataset page: https://huggingface.co/datasets/loganbolton/sketchvlm-connect-dots.

sourceHugging Facecc-by-nc-4.0updated 5mo agoView on Hugging Face

1likes80downloads

Dataset Card

SketchVLM: Physics Ball Drop Dataset

This dataset is part of the SketchVLM framework, which enables Vision-Language Models (VLMs) to produce non-destructive, editable SVG overlays on input images to visually explain their reasoning. The Physics Ball Drop benchmark specifically evaluates a model's ability to predict the trajectory of a ball through obstacles (based on PHYRE).

—Paper: SketchVLM: Vision language models can annotate images to explain thoughts and guide users
—Project Page: https://sketchvlm.github.io/
—GitHub Repository: https://github.com/Brandon-Collins7/sketchvlm

Sample Usage

To download the dataset using the Hugging Face CLI, run:

bash

huggingface-cli download loganbolton/sketchvlm-physics-ball-drop --repo-type dataset --local-dir datasets/ball_drop

Citation

bibtex

@misc{collins2026sketchvlmvisionlanguagemodels,
      title={SketchVLM: Vision language models can annotate images to explain thoughts and guide users},
      author={Brandon Collins and Logan Bolton and Hung Huy Nguyen and Mohammad Reza Taesiri and Trung Bui and Anh Totti Nguyen},
      year={2026},
      eprint={2604.22875},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2604.22875},
}