Provenance graph construction on 3DGS provenance graph dataset 1.0 (test)
1VOResNet+MI
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ResNet+MIFeature extractor=ResNet, Forensic strategy=Mutual Information (MI)2026.06 | 1 | 0.205 | 0.627 | |
| DenseNet+MIFeature extractor=DenseNet, Forensic strategy=Mutual Information (MI)2026.06 | 1 | 0.212 | 0.629 | |
| ViT-B+MIFeature extractor=ViT-B, Forensic strategy=Mutual Information (MI)2026.06 | 1 | 0.23 | 0.639 | |
| ResNet+IntegrityFeature extractor=ResNet, Forensic strategy=Integrity score-based2026.06 | 1 | 0.297 | 0.669 | |
| DenseNet+IntegrityFeature extractor=DenseNet, Forensic strategy=Integrity score-based2026.06 | 1 | 0.319 | 0.68 | |
| ViT-B+IntegrityFeature extractor=ViT-B, Forensic strategy=Integrity score-based2026.06 | 1 | 0.47 | 0.751 | |
| GaussTrace w/o CoTChain-of-thought=false, LLM engine=GPT-52026.06 | 1 | 0.811 | 0.913 | |
| Proposed GaussTraceChain-of-thought=true, LLM engine=GPT-52026.06 | 1 | 0.89 | 0.948 | |
| Rule-based searchForensic strategy=Rule-based search (RBS)2026.06 | 0.939 | 0.319 | 0.566 | |
| Points CD+MIFeature extractor=Points CD (Chamfer Distance), Forensic strategy=Mutual Information (MI)2026.06 | 0.758 | 0.268 | 0.531 | |
| Points CD+IntegrityFeature extractor=Points CD (Chamfer Distance), Forensic strategy=Integrity score-based2026.06 | 0.758 | 0.27 | 0.532 |