Image Captioning on COCO (test) (BLEU@4, METEOR, CIDEr, SPICE)
149.2CIDErPaLI
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| PaLITraining Data=1.6B, Parameters=17B, Backbone/LLM=mT5-XXL2023.11 | 149.2 | — | — | — | |
| CogVLMTraining Data=1.5B, Parameters=6.5B, Backbone/LLM=Vicuna-7B2023.11 | 149.1 | — | — | — | |
| Qwen-VLTraining Data=1.4B, Parameters=9.6B, Backbone/LLM=Qwen-7B2023.11 | 148.7 | — | — | — | |
| REVEALTraining Data=1.3B, Parameters=2.1B, Memory Bank=true, Backbone/LLM=T52023.11 | 145.4 | — | — | — | |
| BLIP-2Training Data=129M, Parameters=1.2B, Backbone/LLM=FlanT5-XL2023.11 | 144.5 | 42.4 | — | — | |
| EVCAPTraining Data=COCO, Parameters=3.97M, Memory Bank=true, Backbone/LLM=Vicuna-13B2023.11 | 140.1 | 41.5 | 31.2 | 24.7 | |
| BLIPTraining Data=129M, Parameters=446M2023.11 | 136.7 | 40.4 | — | — | |
| Qwen3-VL-Instruct RGBPScale=8B2026.04 | 136.6 | 38.18 | 28.9 | 22.24 | |
| MiniGPT4Training Data=5M, Parameters=3.94M, Backbone/LLM=Vicuna-13B2023.11 | 129.6 | 38 | 29.6 | 23.4 | |
| VinVLTraining Data=8.9M, Parameters=110M2023.11 | 129.3 | 38.2 | 30.3 | 23.6 | |
| RCA-NOCTraining Data=COCO, Parameters=110M, Memory Bank=true2023.11 | 128.4 | 37.4 | 29.6 | 23.1 | |
| OSCARTraining Data=4.1M, Parameters=338M2023.11 | 127.8 | 37.4 | 30.7 | 23.5 | |
| AoANet+MATraining Data=COCO, Memory Bank=true2023.11 | 121 | 38 | 28.7 | 21.8 | |
| SmallCapTraining Data=COCO, Parameters=7M, Memory Bank=true, Backbone/LLM=GPT22023.11 | 119.7 | 37 | 27.9 | 21.3 | |
| ClipCapTraining Data=COCO, Parameters=43M, Backbone/LLM=GPT22023.11 | 113.1 | 33.5 | 27.5 | 21.1 | |
| ViECapTraining Data=COCO, Parameters=110M, Backbone/LLM=GPT22023.11 | 92.9 | 27.2 | 24.8 | 18.2 | |
| ProCapBackbone=Vicuna-1.5-7B2026.04 | 86.26 | 26.2 | 22.08 | 14.88 | |
| ProCapBackbone=OpenLLaMA-3B2026.04 | 66.03 | 19.9 | 21.22 | 14.9 | |
| ProCapBackbone=OPT-2.7B2026.04 | 63.4 | 19.7 | 17.78 | 10.96 | |
| ProCapBackbone=TinyLlama-1.1B2026.04 | 46.9 | 13.13 | 17.92 | 12.6 | |
| Qwen3-VL-InstructScale=8B2026.04 | 14.34 | 7.52 | 20.58 | 14.68 | |
| Qwen3-VL-InstructScale=2B2026.04 | 14.1 | 10.4 | 22.18 | 16.14 | |
| Qwen3-VL-InstructScale=4B2026.04 | 12.86 | 6.82 | 20.44 | 14.32 | |
| FastVLMScale=0.5B2026.04 | 11.54 | 6.28 | 16.28 | 9.38 | |
| FastVLMScale=1.5B2026.04 | 10 | 6.2 | 15.38 | 8.92 | |
| FastVLMScale=7B2026.04 | 9.7 | 6.44 | 16.36 | 9.38 |