Research map / Detection and segmentation / Open-set and few-shot detection
Playing for Data: Ground Truth from Computer Games
Stephan R. Richter, Vibhav Vineet, Stefan Roth, Vladlen Koltun
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add 25K synthetic game images to full real training set (ablation) | segmentation model · CamVid | mean IoU: 65.0 -> 68.9 | Improved |
| add synthetic images to 33% of real CamVid training data vs full real only (ablation) | segmentation model · CamVid | mean IoU: 65.0 -> 65.2 | Improved |
| add synthetic images to 25% of real training data vs full real only (ablation) | segmentation model · CamVid | mean IoU: 65.0 -> 63.9 | Hurt |
| add synthetic images to 50% of real training data (ablation) | segmentation model · CamVid | mean IoU: 65.0 -> 66.5 | Improved |
| train on synthetic only versus full real only (ablation) | segmentation model · CamVid | mean IoU: 65.0 -> 43.6 | Hurt |
| add synthetic data to real training (ablation) | segmentation model · KITTI | mean IoU: 59 -> 61.6 | Improved |
| add synthetic data to real training (ablation) | segmentation model · KITTI | Pedestrian IoU: 1.0 -> 29.7 | Improved |
| add synthetic data to real training (ablation) | segmentation model · KITTI | Bicyclist IoU: 27.9 -> 22.6 | Hurt |
| add synthetic data to real training (ablation) | segmentation model · KITTI | Tree IoU: 81 -> 80.3 | Hurt |
| use label propagation with resource hashing instead of manual per-image labeling | annotation pipeline · GTA5 25K images | annotation time (sec/image): 3,600 -> 7 | Improved |
| add association rule mining to MTS propagation (ablation) | annotation pipeline · GTA5 25K images | pixel coverage: 73% -> 98.3% | Improved |
| add synthetic data to real training (ablation) | segmentation model · CamVid | Pole IoU: 23.4 -> 29.5 | Improved |
Closest papers by meaning
- Vision-Zero: Scalable VLM Self-Improvement via Strategic Gamified Self-Play (ICLR 2026 · similarity 0.85)
- MineObserver 2.0: A Deep Learning & In-Game Framework for Assessing Natural Language Descriptions of Minecraft Imagery (AAAI 2024 · similarity 0.84)
- GameVerse: Can Vision-Language Models Learn from Video-based Reflection? (ICML 2026 · similarity 0.84)
- Are Large Vision Language Models Good Game Players? (ICLR 2025 · similarity 0.84)
- Synthesizing Multimodal Verifiable Game Data to Boost VLMs' General Reasoning (ICLR 2026 · similarity 0.83)
- Diffusion Models Are Real-Time Game Engines (ICLR 2025 · 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.
