Research map / Vision-language and multimodal / Human and embodied modeling
Semantic Compositional Networks for Visual Captioning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Compose tag-dependent LSTM weights from tag probabilities instead of feeding tags at each step as features | LSTM-RT₂ · COCO | BLEU-4: 0.322 -> 0.330 | Helped |
| Compose tag-dependent LSTM weights from tag probabilities instead of feeding tags at each step as features | LSTM-RT₂ · Flickr30k | BLEU-4: 0.257 -> 0.265 | Helped |
| Compose tag-dependent LSTM weights from tag probabilities instead of feeding tags at each step as features | LSTM-CRT₂ · Youtube2Text | BLEU-4: 0.469 -> 0.502 | Helped |
| Add tags to a standard LSTM that uses visual features alone (ablation) | LSTM-R · COCO | BLEU-4: 0.292 -> 0.312 | Helped |
| Feed the tag vector at every decoding step rather than only at initialization (ablation) | LSTM-RT · COCO | BLEU-4: 0.316 -> 0.322 | Helped |
| Add tags to a standard LSTM that uses visual features alone (ablation) | LSTM-R · Flickr30k | BLEU-4: 0.201 -> 0.235 | Helped |
| Feed the tag vector at every decoding step rather than only at initialization (ablation) | LSTM-RT · Flickr30k | BLEU-4: 0.235 -> 0.257 | Helped |
| Average predictions from five independently initialized SCN-LSTM models (ablation) | SCN-LSTM · COCO | BLEU-4: 0.330 -> 0.341 | Helped |
| Average predictions from five independently initialized SCN-LSTM models (ablation) | SCN-LSTM · Flickr30k | BLEU-4: 0.265 -> 0.288 | Helped |
| Remove visual feature inputs, retaining only tag inputs (ablation) | SCN-LSTM · COCO qualitative examples | caption correctness | Hurt |
Closest papers by meaning
- A Neural Compositional Paradigm for Image Captioning (NeurIPS 2018 · similarity 0.90)
- FINECAPTION: Compositional Image Captioning Focusing on Wherever You Want at Any Granularity (CVPR 2025 · similarity 0.87)
- Learning Visual Composition through Improved Semantic Guidance (CVPR 2025 · similarity 0.86)
- Set Prediction Guided by Semantic Concepts for Diverse Video Captioning (AAAI 2024 · similarity 0.86)
- Benchmarking Large Vision-Language Models via Directed Scene Graph for Comprehensive Image Captioning (CVPR 2025 · similarity 0.86)
- Advancing Compositional Awareness in CLIP with Efficient Fine-Tuning (NeurIPS 2025 · similarity 0.86)
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.