Image Captioning on Flickr30k (test)
98.3CIDErBlended
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| BlendedBackbone=BLIP-22025.11 | 98.3 | 36.5 | 29.5 | 58.3 | — | — | — | 100 | — | — | — | |
| CleanBackbone=BLIP-22025.11 | 96.6 | 35.7 | 29.1 | 57.8 | — | — | — | — | — | — | — | |
| GLaMMZero-shot=true2023.11 | 95.3 | — | — | — | 18.8 | — | — | — | — | — | — | |
| ShadowCastBackbone=BLIP-22025.11 | 95.1 | 35.8 | 29.2 | 57.6 | — | — | — | 82.7 | — | — | — | |
| CogVLMTraining Data=1.5B, Parameters=6.5B, Backbone/LLM=Vicuna-7B2023.11 | 94.9 | — | — | — | — | — | — | — | — | — | — | |
| AnyDoorBackbone=BLIP-22025.11 | 94.5 | 35 | 29.1 | 57.5 | — | — | — | 99.4 | — | — | — | |
| CleanTarget Label=cat2025.11 | 94 | 34.7 | 28.6 | 57.1 | — | — | — | 4 | — | — | — | |
| CleanArchitecture=LLaVA2025.11 | 92.9 | 34.6 | 28.5 | 56.9 | — | — | — | — | — | — | — | |
| BadNetBackbone=BLIP-22025.11 | 92.7 | 34.7 | 29.4 | 57.4 | — | — | — | 92.4 | — | — | — | |
| VLOODBackbone=BLIP-22025.11 | 92.4 | 34.9 | 28 | 56.8 | — | — | — | 100 | — | — | — | |
| CTPBackbone=BLIP-22025.11 | 92.3 | 34.9 | 28.7 | 57 | — | — | — | 100 | — | — | — | |
| BlendedTarget Label=cat2025.11 | 91.6 | 34 | 28.3 | 55.9 | — | — | — | 2.8 | — | — | — | |
| CGUBTarget Label=cat2025.11 | 91.2 | 34.6 | 27.2 | 56 | — | — | — | 70.5 | — | — | — | |
| AnyDoorArchitecture=LLaVA2025.11 | 89.4 | 33.2 | 28.1 | 55.8 | — | — | — | 100 | — | — | — | |
| BadNetTarget Label=cat2025.11 | 88.8 | 34 | 27.9 | 55.7 | — | — | — | 4 | — | — | — | |
| ShadowCastTarget Label=cat2025.11 | 88.3 | 33.3 | 27.8 | 55.7 | — | — | — | 4 | — | — | — | |
| BlendedArchitecture=LLaVA2025.11 | 88 | 33.5 | 28 | 55.5 | — | — | — | 98.5 | — | — | — | |
| AnyDoorTarget Label=cat2025.11 | 87.9 | 33.5 | 27.8 | 55.4 | — | — | — | 4 | — | — | — | |
| BadNetArchitecture=LLaVA2025.11 | 86.5 | 32.5 | 27.8 | 55.3 | — | — | — | 81.8 | — | — | — | |
| ShadowCastArchitecture=LLaVA2025.11 | 86.3 | 32.5 | 27.9 | 55.4 | — | — | — | 85.5 | — | — | — | |
| Qwen-VLTraining Data=1.4B, Parameters=9.6B, Backbone/LLM=Qwen-7B2023.11 | 85.8 | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5 VL-7BParameters=7B2025.12 | 85.8 | — | — | — | — | — | — | — | — | — | — | |
| HAAVSemi-supervised learning (SSL)=true2023.05 | 85.6 | 34.3 | 25.1 | — | 19 | — | — | — | — | — | — | |
| Qwen2.5 VL-7B + PHMBase Model=Qwen2.5 VL-7B, Compression=PHM, Parameters=~5.38B2025.12 | 85.2 | 31 | 25 | 53.6 | — | 78.8 | — | — | — | — | — | |
| EVCAPTraining Data=COCO, Parameters=3.97M, Memory Bank=true, Backbone/LLM=Vicuna-13B2023.11 | 84.4 | — | — | — | 18 | — | — | — | — | — | — | |
| CTP (Ours)Architecture=LLaVA2025.11 | 83.8 | 32.1 | 27.7 | 55.2 | — | — | — | 95.8 | — | — | — | |
| InstructBLIPZero-shot=true2023.11 | 82.8 | — | — | — | — | — | — | — | — | — | — | |
| HAAVSemi-supervised learning (SSL)=false2023.05 | 81.7 | 34.3 | 24.6 | — | 18 | — | — | — | — | — | — | |
| VLOODArchitecture=LLaVA2025.11 | 81.1 | 27.7 | 25.8 | 52.9 | — | — | — | 98.8 | — | — | — | |
| VLOODTarget Label=cat2025.11 | 80 | 30.3 | 25.2 | 52.6 | — | — | — | 2.2 | — | — | — | |
| SOTAEvaluation Protocol=Pre-training & finetuning2022.09 | 79.6 | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-1.5-13BParameters=13B2025.12 | 79.4 | 30.2 | 26.6 | 53.9 | — | 73.6 | — | — | — | — | — | |
| MiniGPT4Training Data=5M, Parameters=3.94M, Backbone/LLM=Vicuna-13B2023.11 | 78.4 | — | — | — | 16.9 | — | — | — | — | — | — | |
| InstructBLIP-7BParameters=7B2025.12 | 78.2 | 30.9 | 24.8 | 53.4 | — | 77.1 | — | — | — | — | — | |
| PerceptionGPT2023.11 | 77.1 | — | — | — | — | — | — | — | — | — | — | |
| InstructBLIP-7B + PHMBase Model=InstructBLIP-7B, Compression=PHM, Parameters=~5.45B2025.12 | 77 | 30.2 | 24.5 | 53 | — | 76 | — | — | — | — | — | |
| InstructBLIP-13BParameters=13B2025.12 | 76.1 | 30.1 | 24.4 | 53 | — | 76.8 | — | — | — | — | — | |
| FlamingoFT=false, Shot=322022.04 | 75.4 | — | — | — | — | — | — | — | — | — | — | |
| FlamingoFT=false, Shot=42022.04 | 75.1 | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-1.5-7BParameters=7B2025.12 | 74.9 | 28.4 | 26.1 | 52.8 | — | 71.7 | — | — | — | — | — | |
| CAPSFUSIONScale=50M, Evaluation Protocol=Zero-shot2023.10 | 74.5 | — | — | — | 16.9 | — | — | — | — | — | — | |
| ShikraModel Size=13B2023.06 | 73.9 | — | — | — | — | — | — | — | — | — | — | |
| Shikra2023.11 | 73.9 | — | — | — | — | — | — | — | — | — | — | |
| Shikra-13BZero-shot=true2023.11 | 73.9 | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-1.5-7B + PHMBase Model=LLaVA-1.5-7B, Compression=PHM, Parameters=~5.15B2025.12 | 73.5 | 27.9 | 25.7 | 52.3 | — | 70.8 | — | — | — | — | — | |
| CAPSFUSIONScale=100M, Evaluation Protocol=Zero-shot2023.10 | 73 | — | — | — | 16.8 | — | — | — | — | — | — | |
| Flamingo-9BFT=false, Shot=322022.04 | 72.8 | — | — | — | — | — | — | — | — | — | — | |
| Flamingo-9BFT=false, Shot=42022.04 | 72.6 | — | — | — | — | — | — | — | — | — | — | |
| Flamingo-3BFT=false, Shot=42022.04 | 72 | — | — | — | — | — | — | — | — | — | — | |
| Flamingo-3BFT=false, Shot=322022.04 | 71.2 | — | — | — | — | — | — | — | — | — | — | |
| ORT2023.05 | 68.8 | 30.1 | 22.8 | — | 16.9 | — | — | — | — | — | — | |
| VLPModel Type=Single, Pre-training strategy=seq2seq only2019.09 | 68.5 | 31.1 | 23 | — | 17.2 | — | — | — | — | — | — | |
| M²2023.05 | 68.4 | 29.8 | 22.4 | — | 16.2 | — | — | — | — | — | — | |
| CAPSFUSIONScale=10M, Evaluation Protocol=Zero-shot2023.10 | 68.4 | — | — | — | 15.9 | — | — | — | — | — | — | |
| LLaVA-1.6-7BParameters=7B2025.12 | 68.4 | 26.6 | 23.2 | 50.3 | — | 69.6 | — | — | — | — | — | |
| Unified VLPModel Type=Single, Pre-training strategy=joint2019.09 | 67.4 | 30.1 | 23 | — | 17 | — | — | — | — | — | — | |
| Fine-tuned Lfullmode=fine-tuned2021.10 | 67.4 | — | — | — | 17 | — | — | — | — | — | — | |
| Pretrained FT SOTAFT=true, Shot=30K2022.04 | 67.4 | — | — | — | — | — | — | — | — | — | — | |
| FlamingoFT=false, Shot=02022.04 | 67.2 | — | — | — | — | — | — | — | — | — | — | |
| FM-80BModel Size=80B2023.06 | 67.2 | — | — | — | — | — | — | — | — | — | — | |
| Kosmos-12023.06 | 67.1 | — | — | — | — | — | — | — | — | — | — | |
| Kosmos-12023.11 | 67.1 | — | — | — | — | — | — | — | — | — | — | |
| Kosmos-1Zero-shot=true2023.11 | 67.1 | — | — | — | 14.5 | — | — | — | — | — | — | |
| MiniCPM-V-3BParameters=3B2025.12 | 66.8 | 25.1 | 27.2 | 51 | — | 68 | — | — | — | — | — | |
| Kosmos-2Zero-shot=true2023.11 | 66.7 | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-1.6-13BParameters=13B2025.12 | 66.6 | 24.2 | 22.2 | 48.8 | — | 65.2 | — | — | — | — | — | |
| Mixing (Raw & Syn.)Scale=100M, Evaluation Protocol=Zero-shot2023.10 | 66.5 | — | — | — | 15 | — | — | — | — | — | — | |
| Mixing (Raw & Syn.)Scale=50M, Evaluation Protocol=Zero-shot2023.10 | 64.6 | — | — | — | 15.1 | — | — | — | — | — | — | |
| Mixing (Raw & Syn.)Scale=10M, Evaluation Protocol=Zero-shot2023.10 | 64 | — | — | — | 15.1 | — | — | — | — | — | — | |
| SyntheticScale=10M, Evaluation Protocol=Zero-shot2023.10 | 63.9 | — | — | — | 15 | — | — | — | — | — | — | |
| VLPModel Type=Single, Pre-training strategy=bidirectional only2019.09 | 63.3 | 30.5 | 22.6 | — | 16.9 | — | — | — | — | — | — | |
| Up-Down2023.05 | 63.3 | 28.3 | 21.6 | — | 15.9 | — | — | — | — | — | — | |
| GVDModel Type=Single, Input features=with BBox2019.09 | 62.3 | 27.3 | 22.5 | — | 16.5 | — | — | — | — | — | — | |
| RawScale=100M, Evaluation Protocol=Zero-shot2023.10 | 62.2 | — | — | — | 14.6 | — | — | — | — | — | — | |
| Flamingo-9BFT=false, Shot=02022.04 | 61.5 | — | — | — | — | — | — | — | — | — | — | |
| FM-9BModel Size=9B2023.06 | 61.5 | — | — | — | — | — | — | — | — | — | — | |
| Flamingo-3BFT=false, Shot=02022.04 | 60.6 | — | — | — | — | — | — | — | — | — | — | |
| SMALLCAPEvaluation Protocol=Zero-shot (trained on COCO), Datastore=In-domain + Web2022.09 | 60.6 | — | — | — | — | — | — | — | — | — | — | |
| SmallCapTraining Data=COCO, Parameters=7M, Memory Bank=true, Backbone/LLM=GPT22023.11 | 60.6 | — | — | — | — | — | — | — | — | — | — | |
| RawScale=50M, Evaluation Protocol=Zero-shot2023.10 | 60.3 | — | — | — | 14.3 | — | — | — | — | — | — | |
| SyntheticScale=50M, Evaluation Protocol=Zero-shot2023.10 | 60.2 | — | — | — | 14.3 | — | — | — | — | — | — | |
| GVDModel Type=Single2019.09 | 60.1 | 26.9 | 22.1 | — | 16.1 | — | — | — | — | — | — | |
| NBTModel Type=Single, Input features=with BBox2019.09 | 57.5 | 27.1 | 21.7 | — | 15.6 | — | — | — | — | — | — | |
| SyntheticScale=100M, Evaluation Protocol=Zero-shot2023.10 | 57.4 | — | — | — | 13.7 | — | — | — | — | — | — | |
| VLPModel Type=Single, Pre-training strategy=none (baseline)2019.09 | 56.8 | 27.6 | 20.9 | — | 15.3 | — | — | — | — | — | — | |
| DeCap2025.12 | 56.7 | 21.2 | 21.8 | — | — | — | — | — | — | — | — | |
| BUTDModel Type=Single2019.09 | 56.6 | 27.3 | 21.7 | — | 16 | — | — | — | — | — | — | |
| CaMELEvaluation Protocol=Zero-shot (trained on COCO)2022.09 | 55.2 | — | — | — | — | — | — | — | — | — | — | |
| RawScale=10M, Evaluation Protocol=Zero-shot2023.10 | 54.4 | — | — | — | 13.6 | — | — | — | — | — | — | |
| Mixing (Raw & LR)Scale=10M, Evaluation Protocol=Zero-shot2023.10 | 51.7 | — | — | — | 12.2 | — | — | — | — | — | — | |
| Show, Attend & Tell2023.05 | 49.1 | 23.6 | 19.2 | — | 13.3 | — | — | — | — | — | — | |
| Language RewritesScale=10M, Evaluation Protocol=Zero-shot2023.10 | 49 | — | — | — | 11.6 | — | — | — | — | — | — | |
| ViECapTraining Data=COCO, Parameters=110M, Backbone/LLM=GPT22023.11 | 47.9 | — | — | — | 13.6 | — | — | — | — | — | — | |
| Saliency and Context Attention2017.06 | 46.4 | 21.3 | 20 | 45.2 | — | — | — | — | — | — | — | |
| METALMlearning_mode=zero-shot, model_scale=base2022.06 | 43.3 | — | — | — | 11.7 | — | — | — | — | — | — | |
| VISEModel Scale=Qwen3-VL-2B-Instruct2026.06 | 42.64 | — | 26.77 | 47.55 | — | — | — | — | — | — | — | |
| Show & Tell2023.05 | 41.7 | 21.5 | 18.3 | — | 12.2 | — | — | — | — | — | — | |
| ClipCapEvaluation Protocol=Zero-shot (trained on COCO)2022.09 | 41.2 | — | — | — | — | — | — | — | — | — | — | |
| VISEModel Scale=Qwen3-VL-32B-Instruct2026.06 | 40.32 | — | 30.89 | 52.02 | — | — | — | — | — | — | — | |
| VisionZero-RWModel Scale=Qwen3-VL-32B-Instruct2026.06 | 40.05 | — | 30.76 | 51.71 | — | — | — | — | — | — | — |