Research map / Graph learning / Knowledge and temporal graphs
OSCAR-Net: Object-Centric Scene Graph Attention for Image Attribution
Eric Nguyen, Tu Bui, Viswanathan Swaminathan, John Collomosse
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 pretraining of the object CNN f^v (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.2715 -> 0.6282 | Improved |
| add geometry features g_i (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.6282 -> 0.7028 | Improved |
| add shape features s_i (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.7028 -> 0.7015 | No clear effect |
| add relation stream with edge features e_ij (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.7015 -> 0.7148 | Improved |
| add global CNN branch F_G (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.7148 -> 0.7411 | Improved |
| add self-attention pooling instead of first-element embedding (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.6597 -> 0.7411 | Improved |
| add Transformer instead of MLP layers (GNN) (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.7111 -> 0.7411 | Improved |
| use fully connected graph instead of distance-thresholded graph (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.7322 -> 0.7411 | Improved |
| use SimCLR+ content loss instead of SimCLR (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.7013 -> 0.7411 | Improved |
| use SimCLR+ content loss instead of pairwise contrastive loss (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.2199 -> 0.7411 | Improved |
| use SimCLR+ instead of Triplet+ (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.7384 -> 0.7411 | Improved |
| use direct sign() optimization instead of tanh() relaxation (ablation) | OSCAR-Net · Benchmark I | mmAP: 0.7096 -> 0.7411 | Improved |
Closest papers by meaning
- Learning Visual Commonsense for Robust Scene Graph Generation (ECCV 2020 · similarity 0.84)
- Can We Build Scene Graphs, Not Classify Them? FlowSG: Progressive Image-Conditioned Scene Graph Generation with Flow Matching (CVPR 2026 · similarity 0.84)
- SCENIR: Visual Semantic Clarity through Unsupervised Scene Graph Retrieval (ICML 2025 · similarity 0.84)
- Sketching Image Gist: Human-Mimetic Hierarchical Scene Graph Generation (ECCV 2020 · similarity 0.84)
- Spotting the Unseen: Reciprocal Consensus Network Guided by Visual Archetypes (AAAI 2024 · similarity 0.83)
- An Asymmetric Augmented Self-Supervised Learning Method for Unsupervised Fine-Grained Image Hashing (CVPR 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.
