Radiology Report Generation on CheXpert Plus (test)
0.377PrecisionMambaXray-VL-L
Evaluation results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MambaXray-VL-LEncoder=MambaXray-VL, Decoder=Llama2, Time (min)=55.18, Param (M)=202.322024.10 | 0.377 | — | — | — | 0.112 | — | 0.276 | 0.157 | 0.139 | 0.319 | 0.335 | — | — | — | — | — | |
| MambaXray-LVenue=CVPR25, Train Size=1.27M2025.09 | 0.377 | — | — | — | — | — | — | — | — | 0.319 | 0.335 | — | — | — | — | — | |
| CheXagentVenue=arXiv24, Train Size=8.5M2025.09 | 0.373 | — | — | — | — | — | — | — | — | 0.194 | 0.186 | — | — | — | — | — | |
| GPT2-MediumYear=2019, Parameters=354M, Is VLM=false, Training Time (min)=57.332024.10 | 0.358 | — | — | — | 0.063 | — | 0.198 | 0.104 | 0.067 | 0.186 | 0.165 | — | — | — | — | — | |
| GPT2-MediumYear=2019, Params=354M, FT-Size=223K2025.09 | 0.358 | — | — | — | — | — | — | — | — | 0.186 | 0.165 | — | — | — | — | — | |
| TIMEREncoder=Transformer, Decoder=Transformer, Time (min)=26.5, Param (M)=79.282024.10 | 0.345 | — | — | — | 0.083 | — | 0.254 | 0.121 | 0.104 | 0.238 | 0.234 | — | — | — | — | — | |
| TIMERVenue=CHIL23, Train Size=223K2025.09 | 0.345 | — | — | — | — | — | — | — | — | 0.238 | 0.234 | — | — | — | — | — | |
| VLCIEncoder=Transformer, Decoder=Transformer, Time (min)=123.71, Param (M)=91.462024.10 | 0.341 | — | — | — | 0.08 | — | 0.247 | 0.114 | 0.072 | 0.175 | 0.163 | — | — | — | — | — | |
| VLCIVenue=IEEE TIP25, Train Size=223K2025.09 | 0.341 | — | — | — | — | — | — | — | — | 0.175 | 0.163 | — | — | — | — | — | |
| Deepseek-LLMYear=2024, Parameters=6.9B, Is VLM=false, Training Time (min)=201.32024.10 | 0.336 | — | — | — | 0.096 | — | 0.268 | 0.137 | 0.15 | 0.256 | 0.253 | — | — | — | — | — | |
| Yi-1.5Year=2024, Parameters=8.8B, Is VLM=false, Training Time (min)=48.52024.10 | 0.336 | — | — | — | 0.096 | — | 0.269 | 0.138 | 0.155 | 0.241 | 0.243 | — | — | — | — | — | |
| Deepseek-LLMYear=2024, Params=6.9B, FT-Size=223K2025.09 | 0.336 | — | — | — | — | — | — | — | — | 0.256 | 0.253 | — | — | — | — | — | |
| Yi-1.5 (9B)Year=2024, Params=8.8B, FT-Size=223K2025.09 | 0.336 | — | — | — | — | — | — | — | — | 0.241 | 0.243 | — | — | — | — | — | |
| WCLEncoder=Transformer, Decoder=Transformer, Time (min)=24.08, Param (M)=81.292024.10 | 0.335 | — | — | — | 0.084 | — | 0.253 | 0.126 | 0.103 | 0.259 | 0.256 | — | — | — | — | — | |
| WCLVenue=EMNLP21, Train Size=223K2025.09 | 0.335 | — | — | — | — | — | — | — | — | 0.259 | 0.256 | — | — | — | — | — | |
| Vicuna-V1.5Year=2023, Parameters=6.7B, Is VLM=false, Training Time (min)=722024.10 | 0.334 | — | — | — | 0.104 | — | 0.272 | 0.16 | 0.202 | 0.258 | 0.276 | — | — | — | — | — | |
| Vicuna-V1.5Year=2023, Params=6.7B, FT-Size=223K2025.09 | 0.334 | — | — | — | — | — | — | — | — | 0.258 | 0.276 | — | — | — | — | — | |
| MambaXray-VL-BEncoder=MambaXray-VL, Decoder=Llama2, Time (min)=50.66, Param (M)=57.312024.10 | 0.333 | — | — | — | 0.105 | — | 0.267 | 0.149 | 0.117 | 0.264 | 0.273 | — | — | — | — | — | |
| MAMBAXRAY-VL-BPUBLISH=CVPR25, DECODER=LLAMA2-7B2026.02 | 0.333 | — | — | — | — | — | 0.267 | 0.149 | 0.117 | 0.264 | 0.273 | — | — | — | — | — | |
| MambaXray-BVenue=CVPR25, Train Size=1.27M2025.09 | 0.333 | — | — | — | — | — | — | — | — | 0.264 | 0.273 | — | — | — | — | — | |
| Orca-2Year=2023, Parameters=6.7B, Is VLM=false, Training Time (min)=177.332024.10 | 0.33 | — | — | — | 0.103 | — | 0.27 | 0.161 | 0.199 | 0.251 | 0.271 | — | — | — | — | — | |
| Orca-2 (7B)Year=2023, Params=6.7B, FT-Size=223K2025.09 | 0.33 | — | — | — | — | — | — | — | — | 0.251 | 0.271 | — | — | — | — | — | |
| R2GenCMNEncoder=Transformer, Decoder=Transformer, Time (min)=66.08, Param (M)=67.72024.10 | 0.329 | — | — | — | 0.087 | — | 0.256 | 0.127 | 0.102 | 0.241 | 0.231 | — | — | — | — | — | |
| R2GENCMNPUBLISH=ACL21, DECODER=TRANSFORMER2026.02 | 0.329 | — | — | — | — | — | 0.256 | 0.127 | 0.102 | 0.241 | 0.231 | — | — | — | — | — | |
| R2GenCMNVenue=ACL21, Train Size=223K2025.09 | 0.329 | — | — | — | — | — | — | — | — | 0.241 | 0.231 | — | — | — | — | — | |
| CAMANetEncoder=Swin-Former, Decoder=Transformer, Time (min)=23.08, Param (M)=43.222024.10 | 0.328 | — | — | — | 0.083 | — | 0.249 | 0.118 | 0.09 | 0.224 | 0.216 | — | — | — | — | — | |
| CAMANetVenue=IEEE JBHI23, Train Size=223K2025.09 | 0.328 | — | — | — | — | — | — | — | — | 0.224 | 0.216 | — | — | — | — | — | |
| Yi-1.5Year=2024, Parameters=6.1B, Is VLM=false, Training Time (min)=43.662024.10 | 0.322 | — | — | — | 0.091 | — | 0.263 | 0.131 | 0.136 | 0.229 | 0.226 | — | — | — | — | — | |
| Yi-1.5 (6B)Year=2024, Params=6.1B, FT-Size=223K2025.09 | 0.322 | — | — | — | — | — | — | — | — | 0.229 | 0.226 | — | — | — | — | — | |
| Llama-2Year=2023, Parameters=13.0B, Is VLM=false, Training Time (min)=116.422024.10 | 0.321 | — | — | — | 0.101 | — | 0.269 | 0.16 | 0.214 | 0.254 | 0.267 | — | — | — | — | — | |
| Llama-2 (13B)Year=2023, Params=13.0B, FT-Size=223K2025.09 | 0.321 | — | — | — | — | — | — | — | — | 0.254 | 0.267 | — | — | — | — | — | |
| LAYER-WISE EXPERT-ALIGNED DECODING (LEAD)PUBLISH=OURS, DECODER=QWEN3-8B2026.02 | 0.32 | — | — | — | — | — | 0.197 | 0.125 | 0.082 | 0.267 | 0.275 | — | — | — | — | — | |
| R2GenEncoder=Transformer, Decoder=Transformer, Time (min)=110.05, Param (M)=83.52024.10 | 0.318 | — | — | — | 0.081 | — | 0.246 | 0.113 | 0.077 | 0.2 | 0.181 | — | — | — | — | — | |
| R2GENPUBLISH=EMNLP20, DECODER=TRANSFORMER2026.02 | 0.318 | — | — | — | — | — | 0.246 | 0.113 | 0.077 | 0.2 | 0.181 | — | — | — | — | — | |
| R2GenVenue=EMNLP20, Train Size=223K2025.09 | 0.318 | — | — | — | — | — | — | — | — | 0.2 | 0.181 | — | — | — | — | — | |
| Orca-2Year=2023, Parameters=13.0B, Is VLM=false, Training Time (min)=108.662024.10 | 0.317 | — | — | — | 0.1 | — | 0.266 | 0.159 | 0.187 | 0.242 | 0.257 | — | — | — | — | — | |
| Orca-2 (13B)Year=2023, Params=13.0B, FT-Size=223K2025.09 | 0.317 | — | — | — | — | — | — | — | — | 0.242 | 0.257 | — | — | — | — | — | |
| Llama-2Year=2023, Parameters=6.7B, Is VLM=false, Training Time (min)=77.782024.10 | 0.315 | — | — | — | 0.102 | — | 0.267 | 0.157 | 0.179 | 0.244 | 0.26 | — | — | — | — | — | |
| R2GenGPTEncoder=Swin-Transformer, Decoder=Llama2, Time (min)=77.8, Param (M)=90.92024.10 | 0.315 | — | — | — | 0.101 | — | 0.266 | 0.145 | 0.123 | 0.244 | 0.26 | — | — | — | — | — | |
| R2GenCSREncoder=VMamba, Decoder=Llama2, Time (min)=31.2, Param (M)=91.72024.10 | 0.315 | — | — | — | 0.1 | — | 0.265 | 0.146 | 0.121 | 0.247 | 0.259 | — | — | — | — | — | |
| R2GENGPTPUBLISH=META-RAD.23, DECODER=LLAMA2-7B2026.02 | 0.315 | — | — | — | — | — | 0.266 | 0.145 | 0.123 | 0.244 | 0.26 | — | — | — | — | — | |
| R2GENCSRPUBLISH=ARXIV24, DECODER=LLAMA2-7B2026.02 | 0.315 | — | — | — | — | — | 0.265 | 0.146 | 0.121 | 0.247 | 0.259 | — | — | — | — | — | |
| Llama-2 (7B)Year=2023, Params=6.7B, FT-Size=223K2025.09 | 0.315 | — | — | — | — | — | — | — | — | 0.244 | 0.26 | — | — | — | — | — | |
| R2GenGPTVenue=Meta-Rad.23, Train Size=223K2025.09 | 0.315 | — | — | — | — | — | — | — | — | 0.244 | 0.26 | — | — | — | — | — | |
| R2GenCSRVenue=arXiv24, Train Size=223K2025.09 | 0.315 | — | — | — | — | — | — | — | — | 0.247 | 0.259 | — | — | — | — | — | |
| XProNetEncoder=Transformer, Decoder=Transformer, Time (min)=6.3, Param (M)=62.352024.10 | 0.314 | — | — | — | 0.1 | — | 0.265 | 0.146 | 0.121 | 0.247 | 0.259 | — | — | — | — | — | |
| XProNetVenue=ECCV22, Train Size=223K2025.09 | 0.314 | — | — | — | — | — | — | — | — | 0.247 | 0.259 | — | — | — | — | — | |
| Qwen-2Year=2024, Parameters=7.6B, Is VLM=false, Training Time (min)=103.332024.10 | 0.313 | — | — | — | 0.1 | — | 0.27 | 0.142 | 0.159 | 0.269 | 0.261 | — | — | — | — | — | |
| Qwen-2Year=2024, Params=7.6B, FT-Size=223K2025.09 | 0.313 | — | — | — | — | — | — | — | — | 0.269 | 0.261 | — | — | — | — | — | |
| Token-MixerEncoder=ResNet-50, Decoder=Transformer, Time (min)=17.54, Param (M)=104.342024.10 | 0.309 | — | — | — | 0.091 | — | 0.261 | 0.135 | 0.098 | 0.27 | 0.288 | — | — | — | — | — | |
| Token-MixerVenue=IEEE TMI23, Train Size=223K2025.09 | 0.309 | — | — | — | — | — | — | — | — | 0.27 | 0.288 | — | — | — | — | — | |
| InternLMYear=2024, Parameters=7.3B, Is VLM=false, Training Time (min)=2942024.10 | 0.307 | — | — | — | 0.063 | — | 0.207 | 0.136 | 0.104 | 0.274 | 0.284 | — | — | — | — | — | |
| InternLMYear=2024, Params=7.3B, FT-Size=223K2025.09 | 0.307 | — | — | — | — | — | — | — | — | 0.274 | 0.284 | — | — | — | — | — | |
| Llama-3Year=2024, Parameters=8.0B, Is VLM=false, Training Time (min)=1302024.10 | 0.306 | — | — | — | 0.077 | — | 0.22 | 0.121 | 0.134 | 0.232 | 0.222 | — | — | — | — | — | |
| R2GENGPTPUBLISH=META-RAD.23, DECODER=LLAMA3-8B2026.02 | 0.306 | — | — | — | — | — | 0.22 | 0.121 | 0.134 | 0.232 | 0.222 | — | — | — | — | — | |
| Llama-3Year=2024, Params=8.0B, FT-Size=223K2025.09 | 0.306 | — | — | — | — | — | — | — | — | 0.232 | 0.222 | — | — | — | — | — | |
| Qwen-1.5Year=2024, Parameters=7.7B, Is VLM=false, Training Time (min)=154.252024.10 | 0.303 | — | — | — | 0.098 | — | 0.262 | 0.139 | 0.139 | 0.233 | 0.241 | — | — | — | — | — | |
| Qwen-1.5Year=2024, Params=7.7B, FT-Size=223K2025.09 | 0.303 | — | — | — | — | — | — | — | — | 0.233 | 0.241 | — | — | — | — | — | |
| LAYER-WISE EXPERT-ALIGNED DECODING (LEAD)PUBLISH=OURS, DECODER=QWEN3-4B2026.02 | 0.296 | — | — | — | — | — | 0.192 | 0.115 | 0.08 | 0.242 | 0.25 | — | — | — | — | — | |
| Llama-3.1Year=2024, Parameters=8.0B, Is VLM=false, Training Time (min)=1102024.10 | 0.295 | — | — | — | 0.075 | — | 0.221 | 0.121 | 0.136 | 0.251 | 0.242 | — | — | — | — | — | |
| Llama-3.1Year=2024, Params=8.0B, FT-Size=223K2025.09 | 0.295 | — | — | — | — | — | — | — | — | 0.251 | 0.242 | — | — | — | — | — | |
| ORGanEncoder=CNN, Decoder=Transformer, Time (min)=46.66, Param (M)=67.52024.10 | 0.288 | — | — | — | 0.086 | — | 0.261 | 0.135 | 0.107 | 0.287 | 0.277 | — | — | — | — | — | |
| ORGanVenue=ACL23, Train Size=223K2025.09 | 0.288 | — | — | — | — | — | — | — | — | 0.287 | 0.277 | — | — | — | — | — | |
| CvT2DistilGPT2Encoder=Transformer, Decoder=GPT2, Time (min)=13.93, Param (M)=1282024.10 | 0.285 | — | — | — | 0.067 | — | 0.238 | 0.118 | 0.101 | 0.252 | 0.246 | — | — | — | — | — | |
| CVT2DISTILGPT2PUBLISH=AIM23, DECODER=GPT22026.02 | 0.285 | — | — | — | — | — | 0.238 | 0.118 | 0.101 | 0.252 | 0.246 | — | — | — | — | — | |
| CvT2DistilGPT2Venue=AIM23, Train Size=223K2025.09 | 0.285 | — | — | — | — | — | — | — | — | 0.252 | 0.246 | — | — | — | — | — | |
| LAYER-WISE EXPERT-ALIGNED DECODING (LEAD)PUBLISH=OURS, DECODER=QWEN3-2B2026.02 | 0.282 | — | — | — | — | — | 0.188 | 0.108 | 0.075 | 0.237 | 0.243 | — | — | — | — | — | |
| PromptMRGEncoder=ResNet-101, Decoder=Bert, Time (min)=108.45, Param (M)=219.922024.10 | 0.258 | — | — | — | 0.095 | — | 0.222 | 0.121 | 0.044 | 0.265 | 0.281 | — | — | — | — | — | |
| PromptMRGVenue=AAAI24, Train Size=223K2025.09 | 0.258 | — | — | — | — | — | — | — | — | 0.265 | 0.281 | — | — | — | — | — | |
| MiniCPM-V2.5Year=2024, Parameters=8.4B, Is VLM=true, Training Time (min)=51.52024.10 | 0.254 | — | — | — | 0.046 | — | 0.177 | 0.085 | 0.076 | 0.152 | 0.122 | — | — | — | — | — | |
| MiniCPM-V2.5Year=2024, Params=8.4B, FT-Size=223K2025.09 | 0.254 | — | — | — | — | — | — | — | — | 0.152 | 0.122 | — | — | — | — | — | |
| OraPOYear=–, Params=3B, FT-Size=1K2025.09 | 0.237 | — | — | — | — | — | — | — | — | 0.832 | 0.341 | — | — | — | — | — | |
| OraPOVenue=–, Train Size=1K2025.09 | 0.237 | — | — | — | — | — | — | — | — | 0.832 | 0.341 | — | — | — | — | — | |
| Zhu et al.Encoder=Transformer, Decoder=Transformer, Time (min)=10.03, Param (M)=85.952024.10 | 0.217 | — | — | — | 0.074 | — | 0.235 | 0.128 | 0.078 | 0.308 | 0.205 | — | — | — | — | — | |
| Zhu et al.Venue=MICCAI23, Train Size=223K2025.09 | 0.217 | — | — | — | — | — | — | — | — | 0.308 | 0.205 | — | — | — | — | — | |
| InternVL-2Year=2023, Parameters=8.0B, Is VLM=true, Training Time (min)=108.52024.10 | 0.196 | — | — | — | 0.058 | — | 0.188 | 0.112 | 0.085 | 0.127 | 0.132 | — | — | — | — | — | |
| InternVL-2Year=2023, Params=8.0B, FT-Size=223K2025.09 | 0.196 | — | — | — | — | — | — | — | — | 0.127 | 0.132 | — | — | — | — | — | |
| R2GenRLEncoder=Transformer, Decoder=Transformer, Time (min)=44.33, Param (M)=59.872024.10 | 0.193 | — | — | — | 0.035 | — | 0.186 | 0.101 | 0.012 | 0.229 | 0.196 | — | — | — | — | — | |
| R2GENRLPUBLISH=ACL22, DECODER=TRANSFORMER2026.02 | 0.193 | — | — | — | — | — | 0.186 | 0.101 | 0.012 | 0.229 | 0.196 | — | — | — | — | — | |
| R2GenRLVenue=ACL22, Train Size=223K2025.09 | 0.193 | — | — | — | — | — | — | — | — | 0.229 | 0.196 | — | — | — | — | — | |
| Wang et al.Encoder=ViT, Decoder=Llama2, Time (min)=10.82, Param (M)=358.82024.10 | 0.175 | — | — | — | 0.064 | — | 0.22 | 0.11 | 0.059 | 0.099 | 0.078 | — | — | — | — | — | |
| Wang et al.Venue=VIS INT25, Train Size=223K2025.09 | 0.175 | — | — | — | — | — | — | — | — | 0.099 | 0.078 | — | — | — | — | — | |
| ASGMDEncoder=ResNet-101, Decoder=Transformer, Time (min)=87.37, Param (M)=277.412024.10 | 0.146 | — | — | — | 0.063 | — | 0.22 | 0.094 | 0.044 | 0.108 | 0.055 | — | — | — | — | — | |
| ASGMDVenue=ESWA24, Train Size=223K2025.09 | 0.146 | — | — | — | — | — | — | — | — | 0.108 | 0.055 | — | — | — | — | — | |
| MSATEncoder=ViT-B/16, Decoder=Transformer, Time (min)=5.72, Param (M)=141.12024.10 | 0.044 | — | — | — | 0.036 | — | 0.156 | 0.066 | 0.018 | 0.142 | 0.057 | — | — | — | — | — | |
| M2KTEncoder=CNN, Decoder=Transformer, Time (min)=22.5, Param (M)=69.072024.10 | 0.044 | — | — | — | 0.078 | — | 0.247 | 0.101 | 0.077 | 0.142 | 0.058 | — | — | — | — | — | |
| MSATVenue=MICCAI22, Train Size=223K2025.09 | 0.044 | — | — | — | — | — | — | — | — | 0.142 | 0.057 | — | — | — | — | — | |
| M2KTVenue=MIA21, Train Size=223K2025.09 | 0.044 | — | — | — | — | — | — | — | — | 0.142 | 0.058 | — | — | — | — | — | |
| 32-shot ICLBackbone=QoQ-Med2026.05 | — | 21.06 | — | — | 2.54 | — | 15.87 | — | 2.77 | — | — | — | — | 84.87 | 0.1843 | 0.1309 | |
| 8-shot ICLBackbone=LLaVA-Med2026.05 | — | 23.04 | — | — | 2.48 | — | 15.55 | — | 3.49 | — | — | — | — | 85.12 | 0.3206 | 0.1081 | |
| AM-MRG2026.07 | — | 0.381 | 0.238 | — | 0.109 | — | 0.282 | — | 0.221 | — | — | — | — | — | — | — | |
| CD4LMTPF=×3.61, TPS=98.492026.04 | — | — | — | — | — | — | 49.21 | — | 3.38 | — | — | 55.9 | 43.86 | — | — | — | |
| CogRad2026.07 | — | 0.429 | 0.366 | — | 0.316 | — | 0.48 | — | 0.322 | — | — | — | — | — | — | — | |
| d3LLMTPF=×3.84, TPS=101.642026.04 | — | — | — | — | — | — | 44.04 | — | 2.92 | — | — | 55.39 | 40.35 | — | — | — | |
| DIVEBackbone=QoQ-Med2026.05 | — | 20.37 | — | — | 3.93 | — | 18.76 | — | 4.65 | — | — | — | — | 85.05 | 0.4128 | 0.1718 | |
| DIVEBackbone=LLaVA-Med2026.05 | — | 27.79 | — | — | 5.31 | — | 18.99 | — | 7.49 | — | — | — | — | 85.86 | 0.4357 | 0.1688 | |
| dParallelTPF=×4.42, TPS=110.122026.04 | — | — | — | — | — | — | 42.22 | — | 3.17 | — | — | 44.65 | 36.72 | — | — | — | |
| ECHOBaseTPF=×1.89, TPS=48.862026.04 | — | — | — | — | — | — | 56.9 | — | 4.25 | — | — | 59.12 | 52.64 | — | — | — | |
| ECHOblk4TPF=×4.00, TPS=129.192026.04 | — | — | — | — | — | — | 56.14 | — | 4.2 | — | — | 57.4 | 49.57 | — | — | — | |
| ECHOblk8TPF=×8.00, TPS=274.212026.04 | — | — | — | — | — | — | 55.21 | — | 4.14 | — | — | 56.85 | 51.4 | — | — | — |