Note
This is a Hugging Face dataset. For large datasets, ensure huggingface_hub>=1.1.3 to avoid rate limits. Learn more in the Hugging Face integration docs.
Dataset Card for MapTrace-20k#

This is a FiftyOne dataset with 20000 samples.
Installation#
If you haven’t already, install FiftyOne:
pip install -U fiftyone
Usage#
import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("Voxel51/maptrace_20k")
# Launch the App
session = fo.launch_app(dataset)
Dataset Details#
Dataset Description#
MapTrace is a synthetic dataset for path tracing on maps. The dataset contains annotated paths designed to train vision-language models on route-tracing tasks. Each sample consists of a map image annotated with start (green) and end (red) positions, along with a natural language prompt and ground truth path coordinates.
The maptrace_20k split used here contains paths on stylized maps such as those found in brochures, park directories, or shopping malls.
Curated by: Google
Language(s) (NLP): English
License: CC-BY-4.0
Dataset Sources#
Repository: https://huggingface.co/datasets/google/MapTrace
Uses#
Direct Use#
This dataset is intended for training and evaluating vision-language models on spatial reasoning and path-tracing tasks. Models are expected to interpret map images with marked start/end locations and output coordinate sequences representing valid paths between those points.
Dataset Structure#
Original Schema (Hugging Face)#
The maptrace_20k split contains the following fields:
image: The image bytes of the map, annotated with start and end positionslabel: A string representation of a list of (x, y) coordinate tuples defining the target path (normalized between 0 and 1)input: A natural language prompt asking the model to find the path
FiftyOne Schema#
The FiftyOne dataset converts the original format into the following structure:
Sample Fields:
filepath: Path to the PNG image fileinput(StringField): The natural language prompt describing the taskground_truth(Keypoints): The path represented as keypoints with the following properties:Each keypoint is labeled alphabetically (A, B, C, …, Z, AA, AB, etc.)
Points are normalized coordinates in [0, 1] range
The number of keypoints varies per sample
Dataset-Level Attributes:
default_skeleton: AKeypointSkeletonthat connects sequential keypoints (A→B→C→D…) to visualize the path as a connected polyline in the FiftyOne App
Dataset Creation#
Source Data#
Data Collection and Processing#
The dataset is synthetically generated. Maps are created using text-to-image generation models from natural language map descriptions. Paths are then annotated on these synthetic map images with start positions marked in green and end positions marked in red.
Citation#
BibTeX:
@dataset{maptrace2024,
title={MapTrace: A 2M-Sample Synthetic Dataset for Path Tracing on Maps},
author={Google},
year={2024},
publisher={Hugging Face},
url={https://huggingface.co/datasets/google/MapTrace}
}