Minecraft Fly-over Dataset 🎈

⚠️ Please note: This dataset is a work in progress and may be updated at any time. ⚠️

Overview

This dataset offers a high resolution, video chunks capturing a fly-over of diverse Minecraft terrain.

Resolution: 2048x2048
Frame Rate: 30 fps
Field of View (FOV): 85 degrees
Minecraft Version: 1.20.4
Day/Night Cycle: Enabled (Day and night only, no weather effects)
Camera Orientation: Focused slightly towards the sunrise
Biomes Covered: The video spans a wide variety of biomes
Video Quality: The recording bitrate was set to 20.000kbps but the files may vary from 10.000-20.000 kbps
Dataset Size: The entire dataset is ~75gb of footage, ~17 hours in total

Uninterrupted Recording: In each folder, there is a cut up video (10-30 mins per video), which if you rearrange and merge will give you an uninterrupted recording (this is only per folder).
Unmodified World: The Minecraft world was generated using default settings with no customizations or modded biomes.
Camera Positioning: Command blocks were used to remove obstacles, which may cause visible blocks to disappear or the camera to "clip" through terrain.

This dataset can be used for:

Machine Learning: Train models for image or video generation, terrain analysis, etc.
Cinematic Projects: Use the footage for timelapses, animations, or other creative endeavors.
And more: Use it for backgrounds, anything that you can do with it is allowed.

Size: Due to its high resolution and video length, the dataset is quite large (~75GB).
Stuttering: Slight stutters may occur because of the constant loading and unloading of chunks during the fly-over. These can be minimized or removed with code.
Limited Angle: The video captures the world from a single, consistent angle.
Biome Coverage: While a big chunk of biomes are included, not all biomes may be in the dataset, particularly rarer ones.

Chunks are 16x16 areas that are grouped together and loaded/unloaded

Biomes are like different habitats. Deserts, forests, oceans

Feel free to use this footage in your projects, and if you apply it to machine learning or AI work, I'd love to hear about your results!