JayPeKs/train-images-dmx-light-shows-superresolution

DMX Light-Visualization Super-Resolution Dataset

Training images used in the bachelor thesis "Upscaling and Frame Interpolation Techniques for Boosting Light Rendering" (University of Würzburg). The data supports fine-tuning and evaluation of neural super-resolution models (4× upscaling) on 3D light-visualizer renders.

3D light-visualization software mimics stage conditions and amplifies the features of lighting fixtures. As a result the images contain challenging content for super-resolution models: large dark regions, thin laser strips, gobo-textured light beams, and beams whose intensity degrades with distance from the source. These images were gathered specifically to make pretrained models more susceptible to these characteristics.

All images have a native resolution of 1920 × 1080. They were produced with two tools: the GrandMA 3 lighting console software (on-PC version) and Unreal Engine.

Scene Overview

Folder Structure & Origin

ma/1080p/ — GrandMA 3 (screen-recorded)

Recorded with OBS Studio at 30 FPS, then all unique frames were extracted to PNG with FFmpeg. This route was necessary because no direct image-export path from GrandMA 3 was available within the scope of the thesis; recorded frames may contain mild video-compression artifacts.

• —simple_show_front_mp4 — Simple Show (Simple_Show.show, 10 spot fixtures), frontal view.
• —simple_show_left_mp4 — Simple Show, 90° offset view.
• —simple_show_top_mp4 — Simple Show, top-down view.
• —demo_show_front_mp4 — Demo Show (Demoshow_grandMA3.show, 176 fixtures), frontal view of the first sequence.

ue/1080p/ — Unreal Engine (Movie Render Queue)

Rendered with the Movie Render Queue (MRQ) as independent frames with Temporal Anti-Aliasing (TAA) disabled, giving cleaner images than the screen-recorded MA data.

• —Standard_DMX_Demo_Front — DMX Template (UE default previs template, 47 fixtures), frontal baseline sequence, rendered at 60 FPS as PNG.
• —UE_DMX_Renders — DMX Previs Sample (publicly available on Epic's Fab marketplace, 820 fixtures incl. 6 lasers and 41 pyrotechnic fixtures; DMX sequencer track recorded on a GrandMA console). Animated multi-camera sequence rendered at 60 FPS as PNG.
• —DMXPrevis_60fps_MV_DP — DMX Previs Sample re-rendered at 60 FPS as multilayer `.exr` (ZIP / 1-scanline compression) with Deferred Render Data: per-frame Motion Vectors and Depth Buffer (Linear Scene Depth), exported during the Depth Pass. Used for the motion-vector warping / occlusion-mask experiments.
• —CamTest — custom sequence mimicking free camera movement (no camera cuts), used for motion-analysis testing. <!-- VERIFY: confirm this folder's exact role -->

warp/ — Pregenerated warping-training images

PNG images pregenerated with RLFN and SRGAN to train the motion-analysis convolution layer C(2,1) used for temporal refinement. Generated from the same DMX Previs Sample frames used for fine-tuning.

• —rlfn/x4/…
• —srgan/x4/…

Formats

• —PNG — 8-bit RGB frames (the bulk of the dataset).
• —EXR — multilayer float frames carrying Motion Vector and Depth data (ue/1080p/DMXPrevis_60fps_MV_DP).

Licensing & Attribution

[!WARNING] Please review before reuse. This dataset contains frames derived from third-party content and software: - GrandMA 3 default show files and software (MA Lighting). - Unreal Engine's default DMX Template and the DMX Previs Sample asset (obtained via Epic Games' Fab marketplace). The images are shared for academic reproducibility. Redistribution and reuse are subject to the terms of the respective source software/assets.