Research map / Vision-language and multimodal / Human and embodied modeling
ConZIC: Controllable Zero-Shot Image Captioning by Sampling-Based Polishing
Zequn Zeng, Hao Zhang, Ruiying Lu, Dongsheng Wang, Bo Chen, Zhengjue Wang
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 |
|---|---|---|---|
| use shuffled word-update order instead of sequential order (ablation) | ConZIC on MSCOCO Karpathy test · MSCOCO Karpathy test | Div-2: 0.56 -> 0.87 | Improved |
| use shuffled word-update order instead of sequential order (ablation) | ConZIC on MSCOCO Karpathy test · MSCOCO Karpathy test | CLIP-S: 1.01 -> 0.99 | Hurt |
| use shuffled word-update order instead of sequential order (ablation) | ConZIC on MSCOCO Karpathy test · MSCOCO Karpathy test | B-4: 1.31 -> 1.29 | Hurt |
| replace left-to-right autoregressive generation with bidirectional Gibbs-BERT infilling | zero-shot image captioning · MSCOCO noun infilling | B-1: 0.00 -> 0.37 | Improved |
| replace left-to-right autoregressive generation with bidirectional Gibbs-BERT infilling | zero-shot image captioning · MSCOCO verb infilling | B-1: 0.00 -> 0.25 | Improved |
| apply a predefined POS tag sequence as a control signal (ablation) | ConZIC on MSCOCO · MSCOCO POS control | Acc: 15.54 -> 83.36 | Improved |
| apply a predefined POS tag sequence as a control signal (ablation) | ConZIC on MSCOCO · MSCOCO POS control | CLIP-S: 1.01 -> 0.96 | Hurt |
| add iterative polishing updates (ablation) | ConZIC length-controlled captioning · MSCOCO | CLIPScore | Improved |
| increase the caption length (ablation) | ConZIC length control · MSCOCO | CLIPScore | Improved |
| increase the corrupted-word ratio (ablation) | ConZIC multiple-word infilling · MSCOCO | CLIP-S: 0.83 -> 0.89 -> 0.93 | Improved |
| increase the corrupted-word ratio (ablation) | ConZIC multiple-word infilling · MSCOCO | B-4: 60.69 -> 26.08 -> 8.06 | Hurt |
| use ConZIC rather than ZeroCap | zero-shot image captioning · SketchyCOCO caption | B-1: 27.08 -> 39.61 | 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.
- ZeroCap — ZeroCap: Zero-Shot Image-to-Text Generation for Visual-Semantic Arithmetic (CVPR 2022) · better on 6 of 7 shared results
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
- Learning by Correction: Efficient Tuning Task for Zero-Shot Generative Vision-Language Reasoning (CVPR 2024 · similarity 0.88)
- MeaCap: Memory-Augmented Zero-shot Image Captioning (CVPR 2024 · similarity 0.87)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.87)
- Zero-Shot Image Captioning with Multi-type Entity Representations (AAAI 2025 · similarity 0.86)
- Zero-Shot Robustification of Zero-Shot Models (ICLR 2024 · similarity 0.85)
- CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning (ICLR 2026 · similarity 0.85)
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.
