Research map / Vision-language and multimodal / Cross-modal learning
POGA: Paraphrased and Oppositional Graph Alignment for Fine-Grained Cross-Modal Retrieval
Junfeng Zhang, Zhe Xue, Yuankai Qi, Junping Du, Xiangyang Kong, Yishuo Yan, Amin Beheshti, Jian Yang, Anton van den Hengel, Ming-Hsuan Yang
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 hierarchical node contrastive loss to global loss (ablation) | POGA (CLIP ViT-B/16) · DCI | I2T R@1: 66.58 -> 74.21 | Improved |
| add structural relation alignment loss (ablation) | POGA (CLIP ViT-B/16) · DCI | I2T R@1: 74.21 -> 77.37 | Improved |
| add focal disambiguation loss (ablation) | POGA (CLIP ViT-B/16) · DCI | I2T R@1: 77.37 -> 79.44 | Improved |
| add both paraphrased positives and oppositional negatives (ablation) | POGA (CLIP ViT-B/16) · DCI | I2T R@1: 76.52 -> 79.44 | Improved |
| add paraphrased positives only (ablation) | POGA (CLIP ViT-B/16) · DCI | I2T R@1: 76.52 -> 78.11 | Improved |
| add oppositional negatives only (ablation) | POGA (CLIP ViT-B/16) · DCI | I2T R@1: 76.52 -> 78.34 | Improved |
| set relation loss weight to 0 (ablation) | POGA · DCI | I2T R@1: drop of more than 3 points | Hurt |
| set focal loss weight to 0 (ablation) | POGA · DCI | T2I and I2T R@1: about 1-2 point drop | Hurt |
| raise relation loss weight beyond 0.8 (ablation) | POGA · DCI | Recall | Hurt |
| vary global loss weight in range 0.8 to 1.2 (ablation) | POGA · DCI | Recall | No clear effect |
| graph alignment with multi-granularity losses instead of GOAL | CLIP ViT-L/14 · DOCCI to DCI cross-dataset | T2I R@1: 68.93 -> 81.31 | Improved |
| fine-tune with POGA instead of GOAL | CLIP ViT-B/16 · CIFAR100 zero-shot | accuracy: 59.70 -> 67.16 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FineLIP — FineLIP: Extending CLIP's Reach via Fine-Grained Alignment with Longer Text Inputs (CVPR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Fine-Grained Image-Text Matching by Cross-Modal Hard Aligning Network (CVPR 2023 · similarity 0.91)
- Beyond Global Similarity: Multi-Conditional Retrieval for Fine-Grained Cross-Modal Understanding (CVPR 2026 · similarity 0.89)
- Cross Modal Fine-grained Alignment via Granularity-aware and Region-uncertain Modeling (AAAI 2026 · similarity 0.89)
- Cross-Modal and Uni-Modal Soft-Label Alignment for Image-Text Retrieval (AAAI 2024 · similarity 0.89)
- DiCA: Disambiguated Contrastive Alignment for Cross-Modal Retrieval with Partial Labels (AAAI 2025 · similarity 0.87)
- WikiDO: A New Benchmark Evaluating Cross-Modal Retrieval for Vision-Language Models (NeurIPS 2024 · similarity 0.87)
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.
