Research map / Image generation / Super-resolution
QuickRender: A Photorealistic Procedurally Generated Dataset with Applications to Super Resolution (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces QuickRender, a procedurally generated Blender dataset of over 5,000 sequential image triplets with auxiliary data, and MetaSRGAN, an SRGAN variant that uses auxiliary render data for super resolution.
Problem
Full photorealistic Blender rendering at high resolution is slow, and existing synthetic or real datasets for temporal tasks like optical flow lack scene diversity, photorealism, or auxiliary data.
Approach
A procedurally randomized Blender scene generator produces photorealistic image triplets with segmentation, normal, depth and velocity maps, and MetaSRGAN injects low-resolution auxiliary render data into SRGAN's residual blocks to upscale low-resolution renders faster than full rendering.
Key results
- MetaSRGAN achieved SSIM 0.6958 and PSNR 26.4128 versus SRGAN's 0.6440 and 23.3231.
- MetaSRGAN gave an average 90.52% reduction in render time relative to rendering full-resolution output for 2k and 4k images.
- QuickRender contains 5 procedurally randomized scenes and over 5,000 sequential image triplets with segmentation, normal, depth and velocity maps.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add low-resolution depth, object segmentation maps, and surface normals as auxiliary inputs | SRGAN on QuickRender · QuickRender | SSIM: 0.6440 -> 0.6958 | Helped |
| Add low-resolution depth, object segmentation maps, and surface normals as auxiliary inputs | SRGAN on QuickRender · QuickRender | PSNR: 23.3231 -> 26.4128 | Helped |
| Infer high-resolution images from low-resolution renders with MetaSRGAN | MetaSRGAN · 2k and 4k square images | render time reduction: 90.52% | Helped |
| Add velocity maps as auxiliary inputs (ablation) | MetaSRGAN · Super resolution | final result | No clear effect |
Surprising findings
- Adding velocity maps increased network size without a visible benefit to the super resolution result, so they were excluded.
Limitations
- Only 5 scenes, limiting diversity.
- Evaluation relies solely on SSIM and PSNR, which the authors note may not reflect perceptual quality.
- Super resolution results are preliminary and demonstrated on a single out-of-dataset scene qualitatively.
Experimental setup
Datasets: QuickRender
Benchmarks: QuickRender
Metrics: SSIM PSNR render time reduction
Baselines: SRGAN
Builds on: SRGAN Blender FlyingChairs FlyingThings3D
Closest papers by meaning
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · similarity 0.84)
- 4K4D: Real-Time 4D View Synthesis at 4K Resolution (CVPR 2024 · similarity 0.84)
- DreamScene4D: Dynamic Multi-Object Scene Generation from Monocular Videos (NeurIPS 2024 · similarity 0.83)
- Instant4D: 4D Gaussian Splatting in Minutes (NeurIPS 2025 · similarity 0.83)
- Coarse-to-Real: Generative Rendering for Populated Dynamic Scenes (NeurIPS 2026 · similarity 0.83)
- 4Real: Towards Photorealistic 4D Scene Generation via Video Diffusion Models (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.