PaulineLi/QuantiPhy-validation

QuantiPhy (Validation Set)

Dataset Summary

QuantiPhy is a benchmark for evaluating whether vision–language models (VLMs) can perform quantitative physical inference from visual evidence, rather than producing plausible but ungrounded numerical guesses.

This repository contains the official validation set of QuantiPhy, released to support model development, ablation studies, and preliminary evaluation.
The validation set represents approximately 4% of the full benchmark and consists of 159 video–question–answer (QA) pairs.

Each instance requires a model to output a single continuous numerical value (e.g., object size, velocity, or acceleration) in real-world units, given a short video and a natural-language question.

---

Intended Use

This validation release is intended for:

• model debugging and prompt development,
• hyperparameter tuning,
• ablation and error analysis,
• sanity checks prior to full benchmark evaluation.

It is not intended to be used as a substitute for the full QuantiPhy benchmark.
The complete dataset, including training and test splits, will be released separately.

---

Supported Tasks

• Video-based numerical regression
• Quantitative visual reasoning
• Vision–language model evaluation

Tasks cover three core kinematic properties:

• Size
• Velocity
• Acceleration

All questions are open-ended and require predicting a real-valued scalar.

---

Dataset Structure

Each instance is represented as a structured video–text record with the following fields:

Field	Description
video_id	Unique identifier for the video
video_source	Data source (simulation, lab, or internet)
video_type	Four-character code encoding task configuration
fps	Frame rate of the video
inference_type	Static or dynamic prior/target configuration
question	Natural-language question with explicit physical units
prior	Physical prior provided in world units (e.g., object size, velocity, or acceleration)
depth_info	Depth/distance information for 3D configurations (if applicable)
answer	Ground-truth numerical value (float, real-world units)

Videos are short (typically 2–3 seconds) and recorded with a static camera to ensure well-defined kinematic inference.

---

Task Design Overview

Each instance provides the model with:

• a short video depicting object motion, and
• one physical prior in world units (object size, velocity at a given timestamp, or acceleration at a given timestamp).

The model is then asked to infer a target kinematic quantity—possibly for a different object—expressed in real-world units.

Tasks vary along four axes:

1. Physical prior: Size (S), Velocity (V), Acceleration (A)
2. Dimensionality: 2D (planar motion) or 3D (with depth variation)
3. Object setting: Single-object (S) or multi-object (M)
4. Background complexity: Plain (X), Simple (S), Complex (C)

---

Validation Set Statistics

• 159 QA pairs
• Covers all three physical priors (S / V / A)
• Includes both 2D and 3D configurations
• Videos sourced from:
  • Blender simulations,
  • laboratory captures,
  • curated internet videos

This subset is designed to be representative but non-exhaustive relative to the full benchmark.

---

Data Sources and Quality Control

• Simulation: Blender-rendered scenes with precise physical ground truth.
• Laboratory capture: Real-world recordings using calibrated depth and multi-view setups.
• Internet / author-recorded videos: Carefully curated monocular videos meeting strict physical constraints.

All videos undergo manual review to remove:

• excessive motion blur,
• severe occlusion,
• untrackable motion,
• personally identifiable information (PII).

---

License

The annotations and metadata in this repository are released under the
Creative Commons Attribution 4.0 (CC BY 4.0) license.

Videos originate from simulated environments, laboratory recordings, and publicly available sources.
Each video remains subject to its original license and terms of use.

This release is intended for research and evaluation purposes.

---

Authors

Puyin Li*, Tiange Xiang*, Ella Mao*,
Shirley Wei, Xinye Chen, Adnan Masood,
Li Fei-Fei†, Ehsan Adeli†

* Equal contribution.

---

Citation

If you use this validation set in your work, please cite:

@article{li2025quantiphy,
  title   = {QuantiPhy: A Quantitative Benchmark Evaluating Physical Reasoning Abilities of Vision-Language Models},
  author  = {Li, Puyin and Xiang, Tiange and Mao, Ella and Wei, Shirley and Chen, Xinye and Masood, Adnan and Li, Fei-Fei and Adeli, Ehsan},
  journal = {arXiv preprint arXiv:2512.19526},
  year    = {2025}
}