HUE Dataset: High-Resolution Event and Frame Sequences for Low-Light Vision

Burak Ercan 1 Onur Eker 1,2 Aykut Erdem 3,4 Erkut Erdem 1
1 Hacettepe University, Computer Engineering Department 2 HAVELSAN Inc.
3 Koç University, Computer Engineering Department 4 Koç University, KUIS AI Center
ECCV Workshop on Neuromorphic Vision, 2024

Paper
Supplementary
Publication

Here we present HUE (Hacettepe University Event) Dataset, a comprehensive collection of high-resolution event and frame sequences captured in diverse and challenging low-light conditions. Our dataset offers 106 sequences across 6 categories, recorded using a hybrid RGB and event camera setup with events at 1280×720 and frames at 1456×1088 resolution. The sequences feature both static and dynamic scenes captured in indoor and outdoor environments, at various times of day, with illuminance levels ranging from 0 to 24 lux. Dataset parts are hosted on Zenodo, with download links below.

Indoor
23 sequences · 6 mins · 0–3 lux

Indoor environments dimly lit via natural or artificial light sources.

Download: Part 1, Part 2, Part 3
City
11 sequences · 4 mins · 0–14 lux

Capturing cityscapes through the window of a mid-rise building.

Download: Part 1, Part 2
Twilight
23 sequences · 10 mins · 0–24 lux

Outdoors during twilight, featuring natural and urban elements.

Download: Part 1, Part 2, Part 3, Part 4
Night
16 sequences · 3 mins · ~0 lux

Taken in urban environments during night, with moving people, cars, etc.

Download: HUE-Night
Driving
16 sequences · 6 mins · ~0 lux

From the windshield of a car driving around the city in twilight and night.

Download: Part 1, Part 2
Controlled
17 sequences · 2 mins · 0–2 lux

Same scene captured under varying lighting levels and camera settings.

Download: HUE-Controlled

Additional calibration sequences (flickering chessboard patterns for estimating intrinsic and extrinsic camera parameters)

Dataset Format

The events in the HUE dataset are stored in a numpy memmap format, similar to that of the event_utils library, and compatible with EVREAL. Each sequence folder contains the following files:

File Description
events_p.npy Event polarities.
events_ts.npy Event timestamps in seconds.
events_xy.npy Event (x, y) coordinates.
images_ts.npy Image timestamps (midpoint of exposure).
images_exp_start_ts.npy Image exposure start timestamps.
images_exp_end_ts.npy Image exposure end timestamps.
image_event_indices.npy For each image frame, the index of the last event at or before the image timestamp. Useful for aligning events with frames.
png_images/ 8-bit 3-channel RGB images (1456×1088) captured by the frame camera, saved as PNG files (rgb_00000.png, rgb_00001.png, ...).
lux_values.txt Light level (in lux) measured on the event sensor, recorded with each frame. One value per line, one per frame.
metadata.json Sequence metadata, including event sensor resolution ({"sensor_resolution": [720, 1280]}).
event_camera_params.json Event camera parameters captured via Prophesee SDK (sensor info, biases, firmware version, etc.).
frame_camera_params.xml Frame camera settings captured via Allied Vision Vimba SDK (exposure, gain, frame rate, etc.).

All timestamps are in seconds, starting from 0.

BibTeX 

@inproceedings{ercan2025hue,
title={{HUE} Dataset: High-Resolution Event and Frame Sequences for Low-Light Vision},
author={Ercan, Burak and Eker, Onur and Erdem, Aykut and Erdem, Erkut},
booktitle={Computer Vision -- ECCV 2024 Workshops},
year={2025},
pages={174-191}}

Acknowledgements
This work was supported by TUBITAK-1001 Program Award No. 121E454.