Research map / Vision-language and multimodal / CLIP and zero-shot
Transferable Decoding with Visual Entities for Zero-Shot Image Captioning
Junjie Fei, Teng Wang, Jinrui Zhang, Zhenyu He, Chengjie Wang, Feng Zheng
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 entity-aware hard prompts (with masking) to CapDec-style soft prompts (ablation) | CapDec · COCO => NoCaps val out-of-domain | CIDEr: 28.7 -> 65.0 | Improved |
| add entity masking (40%) to entity hard prompts (ablation) | ViECap · COCO test | CIDEr: 53.2 -> 92.9 | Improved |
| increase entity masking ratio beyond 40% (ablation) | ViECap · NoCaps val overall | CIDEr: 66.2 -> 65.5 -> 65.3 (60%, 80%) | Mixed |
| combine soft and hard prompts instead of entity-only or soft-only (ablation) | ViECap · NoCaps val overall | CIDEr: 45.9 (soft) / 44.4 (entity) -> 66.2 | Improved |
| use CLIP prompt ensemble for entity retrieval (ablation) | ViECap · NoCaps val overall | CIDEr: 61.0 -> 66.2 | Improved |
| scale up frozen language model from GPT-2 Base to OPT-2.7B (ablation) | ViECap · NoCaps val overall | CIDEr: 62.0 -> 72.1 | Improved |
| increase soft prompt length from 10 to 20, 30, 40 (ablation) | ViECap · NoCaps val overall | CIDEr: 66.2 vs 65.6 / 66.0 / 65.9 | No clear effect |
| change hard prompt template wording (ablation) | ViECap · NoCaps val overall | CIDEr: 66.2 vs 65.2 / 65.3 / 64.9 | No clear effect |
| use large noisy Visual Genome vocabulary instead of cleaned VGOI vocabulary (ablation) | ViECap · NoCaps val overall | CIDEr: 66.2 -> 50.0 | Hurt |
| add entity-aware hard prompts under in-domain captioning | CapDec · Flickr30k test | CIDEr: 39.1 -> 47.9 | Improved |
| train with 0.1% of COCO data | ViECap vs CapDec · NoCaps val overall | CIDEr: 10.4 -> 30.2 | Improved |
| train on 10% COCO data | ViECap vs CapDec · COCO test | CIDEr: 83.6 vs 83.4 | No clear effect |
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.
- DeCap — DeCap: Decoding CLIP Latents for Zero-Shot Captioning via Text-Only Training (ICLR 2023) · 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
- Zero-Shot Image Captioning with Multi-type Entity Representations (AAAI 2025 · similarity 0.90)
- FlexCap: Describe Anything in Images in Controllable Detail (NeurIPS 2024 · similarity 0.88)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.88)
- Learning by Correction: Efficient Tuning Task for Zero-Shot Generative Vision-Language Reasoning (CVPR 2024 · similarity 0.87)
- Panoptic Captioning: An Equivalence Bridge for Image and Text (NeurIPS 2025 · similarity 0.87)
- CapsFusion: Rethinking Image-Text Data at Scale (CVPR 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.
