EEG-to-text translation on EEG-ImageNet CVPR2017 (test)
32.7ROUGE-1SYNAPSE
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| SYNAPSEAblation ID=A4 (-Exe), Decoder Model=Gemini 2.5 Flash Lite2026.05 | 32.7 | 8.7 | 28.9 | 27.8 | 5.9 | 27.9 | 90.3 | 4.67 | 1.36 | |
| SYNAPSEAblation ID=A1 (Full), Decoder Model=Gemini 2.5 Flash Lite2026.05 | 32.3 | 8.6 | 28.6 | 27.3 | 5.8 | 27.5 | 90.3 | 4.68 | 1.37 | |
| SYNAPSEAblation ID=A3 (-Obj), Decoder Model=Gemini 2.5 Flash Lite2026.05 | 32.3 | 8.4 | 28.4 | 27.4 | 5.7 | 27.4 | 90.3 | 4.64 | 1.35 | |
| SYNAPSEAblation ID=A3 (-Obj), Decoder Model=Qwen2.5-7B2026.05 | 32.3 | 9.2 | 28.6 | 27.6 | 6.1 | 28.2 | 90.3 | 4.38 | 1.34 | |
| SYNAPSEAblation ID=A5 (-Fac), Decoder Model=Gemini 2.5 Flash Lite2026.05 | 32.3 | 8.4 | 28.3 | 27.8 | 5.6 | 28 | 90.2 | 4.66 | 1.34 | |
| SYNAPSEAblation ID=A1 (Full), Decoder Model=Qwen2.5-7B2026.05 | 32.2 | 9.3 | 28.7 | 27.3 | 6.1 | 28 | 90.4 | 4.48 | 1.35 | |
| SYNAPSEAblation ID=A4 (-Exe), Decoder Model=Qwen2.5-7B2026.05 | 32.2 | 9.3 | 28.8 | 27.3 | 6.2 | 27.9 | 90.4 | 4.46 | 1.35 | |
| SYNAPSEAblation ID=A5 (-Fac), Decoder Model=Qwen2.5-7B2026.05 | 32 | 9.2 | 28.6 | 26.8 | 6.2 | 26.9 | 90.5 | 4.57 | 1.36 | |
| SYNAPSEAblation ID=A1 (Full), Decoder Model=GPT-4o-mini2026.05 | 31.7 | 8.8 | 28.1 | 27.6 | 5.9 | 28.2 | 90.2 | 4.79 | 1.37 | |
| SYNAPSEAblation ID=A2 (m = 0), Decoder Model=Qwen2.5-7B2026.05 | 31.6 | 8.4 | 28 | 26.7 | 5.6 | 26.7 | 90.3 | 4.44 | 1.31 | |
| SYNAPSEAblation ID=A3 (-Obj), Decoder Model=GPT-4o-mini2026.05 | 31.6 | 8.7 | 27.9 | 27.6 | 5.9 | 28.2 | 90.2 | 4.78 | 1.36 | |
| SYNAPSEAblation ID=A4 (-Exe), Decoder Model=GPT-4o-mini2026.05 | 31.6 | 8.9 | 28.1 | 27.6 | 6 | 28.2 | 90.2 | 4.77 | 1.36 | |
| Gemini 2.5 Flash Lite (Focal)Similarity Refiner Loss=Focal, Semantic Anchor=WithObj2026.03 | 31.5 | 8.5 | 28.7 | 25.2 | 5.6 | 26.1 | 89.7 | 4.77 | 1.4 | |
| SENSEDecoder Model=Gemini 2.5 Flash LiteWith Obj2026.05 | 31.5 | 8.5 | 28.7 | 25.2 | 5.6 | 26.1 | 89.7 | 4.77 | 1.4 | |
| SYNAPSEAblation ID=A2 (m = 0), Decoder Model=Gemini 2.5 Flash Lite2026.05 | 31.4 | 7.6 | 27.7 | 26.5 | 5.2 | 25.5 | 90.1 | 4.61 | 1.31 | |
| SYNAPSEAblation ID=A6 (Base), Decoder Model=Gemini 2.5 Flash Lite2026.05 | 31 | 7.5 | 26.9 | 26.6 | 5 | 27.5 | 89.8 | 4.58 | 1.3 | |
| SYNAPSEAblation ID=A5 (-Fac), Decoder Model=GPT-4o-mini2026.05 | 30.9 | 8.2 | 27.3 | 26.9 | 5.5 | 28 | 90 | 4.78 | 1.35 | |
| Gemini 2.5 Flash Lite (Binary Cross Entropy)Similarity Refiner Loss=Binary Cross Entropy, Semantic Anchor=WithObj2026.03 | 30.8 | 7.7 | 27.7 | 24.4 | 5.2 | 25.1 | 89.4 | 4.75 | 1.4 | |
| SYNAPSEAblation ID=A2 (m = 0), Decoder Model=GPT-4o-mini2026.05 | 30.8 | 8.2 | 27.3 | 26.9 | 5.5 | 27 | 90.1 | 4.78 | 1.33 | |
| GPT-4o-mini (Focal)Similarity Refiner Loss=Focal, Semantic Anchor=WithObj2026.03 | 30.6 | 8.2 | 28.2 | 24.9 | 5.6 | 25.3 | 89.8 | 4.75 | 1.4 | |
| SENSEDecoder Model=GPT-4o-miniWith Obj2026.05 | 30.6 | 8.2 | 28.2 | 24.9 | 5.6 | 25.3 | 89.8 | 4.75 | 1.4 | |
| Gemini 2.5 Flash Lite (Focal)Similarity Refiner Loss=Focal, Semantic Anchor=WithoutObj2026.03 | 30.5 | 7.6 | 27.1 | 24.7 | 4.9 | 24.4 | 89.3 | 4.67 | 1.3 | |
| SYNAPSEAblation ID=A3 (-Obj), Decoder Model=LLaMa-3-8B2026.05 | 30.4 | 8.6 | 27.3 | 25.8 | 5.5 | 27.4 | 90.2 | 4.71 | 1.34 | |
| GPT-4o-mini (Focal)Similarity Refiner Loss=Focal, Semantic Anchor=WithoutObj2026.03 | 30.3 | 7.3 | 27.1 | 25.2 | 4.7 | 24.5 | 89.5 | 4.6 | 1.27 | |
| Gemini 2.5 Flash Lite (Contrastive Multi Label)Similarity Refiner Loss=Contrastive Multi Label, Semantic Anchor=WithObj2026.03 | 30.2 | 7.7 | 27.3 | 24.4 | 5.2 | 24.9 | 89.4 | 4.77 | 1.4 | |
| Thought2TextBackbone=LLaMa-3-8B, Prompting=ALL2026.03 | 30 | 8.1 | 26.6 | 25.5 | 5.5 | 26.3 | 89 | 4.82 | 1.58 | |
| SYNAPSEAblation ID=A1 (Full), Decoder Model=LLaMa-3-8B2026.05 | 30 | 8.2 | 27.2 | 25.1 | 5.4 | 26.8 | 90.2 | 4.71 | 1.35 | |
| THOUGHT2TEXTDecoder Model=LLaMa-3-8BALL2026.05 | 30 | 8.1 | 26.6 | 25.5 | 5.5 | 26.3 | 89 | 4.82 | 1.58 | |
| SYNAPSEAblation ID=A4 (-Exe), Decoder Model=LLaMa-3-8B2026.05 | 29.8 | 8.2 | 27.1 | 25 | 5.4 | 26.7 | 90.2 | 4.72 | 1.36 | |
| GPT-4o-mini (Binary Cross Entropy)Similarity Refiner Loss=Binary Cross Entropy, Semantic Anchor=WithObj2026.03 | 29.6 | 7.4 | 27.1 | 23.7 | 5.1 | 24.4 | 89.7 | 4.71 | 1.38 | |
| SYNAPSEAblation ID=A2 (m = 0), Decoder Model=LLaMa-3-8B2026.05 | 29.5 | 7.9 | 26.8 | 24.6 | 5.2 | 26.1 | 90.1 | 4.69 | 1.33 | |
| GPT-4o-mini (Binary Cross Entropy)Similarity Refiner Loss=Binary Cross Entropy, Semantic Anchor=WithoutObj2026.03 | 29.4 | 6.6 | 26.2 | 24.2 | 4.5 | 23.5 | 89.3 | 4.58 | 1.28 | |
| GPT-4o-mini (Contrastive Multi Label)Similarity Refiner Loss=Contrastive Multi Label, Semantic Anchor=WithObj2026.03 | 29.2 | 7.7 | 27 | 23.8 | 5.3 | 24.1 | 89.5 | 4.7 | 1.38 | |
| SYNAPSEAblation ID=A6 (Base), Decoder Model=Qwen2.5-7B2026.05 | 29.2 | 6.6 | 24.9 | 24.9 | 4.4 | 26 | 89.6 | 4.24 | 1.27 | |
| Gemini 2.5 Flash Lite (Binary Cross Entropy)Similarity Refiner Loss=Binary Cross Entropy, Semantic Anchor=WithoutObj2026.03 | 29.1 | 6.2 | 25.9 | 23.4 | 4.4 | 23.1 | 88.9 | 4.72 | 1.31 | |
| SYNAPSEAblation ID=A5 (-Fac), Decoder Model=LLaMa-3-8B2026.05 | 29.1 | 8.1 | 26.6 | 24 | 5.3 | 26.2 | 90.1 | 4.72 | 1.38 | |
| GPT-4o-mini (Contrastive Multi Label)Similarity Refiner Loss=Contrastive Multi Label, Semantic Anchor=WithoutObj2026.03 | 28.3 | 6.2 | 25.5 | 24.1 | 4.4 | 22.8 | 89.1 | 4.64 | 1.29 | |
| SYNAPSEAblation ID=A6 (Base), Decoder Model=GPT-4o-mini2026.05 | 28 | 5.9 | 24.3 | 24.1 | 4.1 | 26.2 | 89.4 | 4.78 | 1.31 | |
| Gemini 2.5 Flash Lite (Contrastive Multi Label)Similarity Refiner Loss=Contrastive Multi Label, Semantic Anchor=WithoutObj2026.03 | 27.7 | 6.2 | 25 | 23 | 4.3 | 22.8 | 88.8 | 4.71 | 1.3 | |
| Thought2TextBackbone=Qwen2.5-7B, Prompting=ALL2026.03 | 26.4 | 4.6 | 22.8 | 22.7 | 3.7 | 21.1 | 88 | 4.75 | 1.28 | |
| THOUGHT2TEXTDecoder Model=Qwen2.5-7BALL2026.05 | 26.4 | 4.6 | 22.8 | 22.7 | 3.7 | 21.1 | 88 | 4.75 | 1.28 | |
| GPT-4o-mini (Naive Baseline)Similarity Refiner Loss=Naive Baseline, Semantic Anchor=WithObj2026.03 | 25.8 | 4.8 | 23.8 | 21.6 | 3.8 | 20.2 | 88.8 | 4.59 | 1.22 | |
| Gemini 2.5 Flash Lite (Naive Baseline)Similarity Refiner Loss=Naive Baseline, Semantic Anchor=WithObj2026.03 | 25.8 | 5.2 | 24.5 | 18.1 | 3.6 | 19.3 | 88.8 | 4.7 | 1.28 | |
| SYNAPSEAblation ID=A6 (Base), Decoder Model=LLaMa-3-8B2026.05 | 25.3 | 5.3 | 21.6 | 21.4 | 3.5 | 24.8 | 89.1 | 4.46 | 1.25 | |
| Gemini 2.5 Flash Lite (Naive Baseline)Similarity Refiner Loss=Naive Baseline, Semantic Anchor=WithoutObj2026.03 | 23.9 | 3.8 | 22.3 | 18.9 | 3.1 | 16.9 | 87.9 | 4.45 | 1.07 | |
| GPT-4o-mini (Naive Baseline)Similarity Refiner Loss=Naive Baseline, Semantic Anchor=WithoutObj2026.03 | 22.7 | 3.1 | 20.7 | 20 | 3.1 | 17.1 | 87.7 | 4.39 | 1.06 | |
| Qwen2.5-7B (Focal)Similarity Refiner Loss=Focal, Semantic Anchor=WithObj2026.03 | 21.8 | 5.9 | 19.9 | 13.8 | 3.2 | 15 | 87.6 | 3.47 | 1.33 | |
| SENSEDecoder Model=Qwen2.5-7BWith Obj2026.05 | 21.8 | 5.9 | 19.9 | 13.8 | 3.2 | 15 | 87.6 | 3.47 | 1.33 | |
| Qwen2.5-7B (Contrastive Multi Label)Similarity Refiner Loss=Contrastive Multi Label, Semantic Anchor=WithObj2026.03 | 20.3 | 5.4 | 18.6 | 12.8 | 3 | 13.8 | 87.3 | 3.27 | 1.34 | |
| Qwen2.5-7B (Focal)Similarity Refiner Loss=Focal, Semantic Anchor=WithoutObj2026.03 | 20.2 | 4.5 | 17.7 | 14.2 | 2.9 | 15.4 | 87.4 | 3.41 | 1.24 | |
| Qwen2.5-7B (Binary Cross Entropy)Similarity Refiner Loss=Binary Cross Entropy, Semantic Anchor=WithObj2026.03 | 20.1 | 5.4 | 18.1 | 12.7 | 3.1 | 13.9 | 87.5 | 3.27 | 1.33 | |
| Qwen2.5-7B (Binary Cross Entropy)Similarity Refiner Loss=Binary Cross Entropy, Semantic Anchor=WithoutObj2026.03 | 18.7 | 4 | 16.3 | 12.8 | 2.6 | 14 | 86.9 | 3.2 | 1.25 | |
| Qwen2.5-7B (Contrastive Multi Label)Similarity Refiner Loss=Contrastive Multi Label, Semantic Anchor=WithoutObj2026.03 | 17.5 | 3.9 | 15.6 | 12 | 2.5 | 12.9 | 86.9 | 3.23 | 1.25 | |
| Qwen2.5-7B (Naive Baseline)Similarity Refiner Loss=Naive Baseline, Semantic Anchor=WithObj2026.03 | 17.1 | 3.6 | 15.9 | 10 | 2.1 | 10.2 | 86.2 | 3.02 | 1.16 | |
| LLaMa-3-8B (Focal)Similarity Refiner Loss=Focal, Semantic Anchor=WithoutObj2026.03 | 15.9 | 2.7 | 14.3 | 10.4 | 2.1 | 14.8 | 88.2 | 4.46 | 1.24 | |
| LLaMa-3-8B (Contrastive Multi Label)Similarity Refiner Loss=Contrastive Multi Label, Semantic Anchor=WithoutObj2026.03 | 15.1 | 2.5 | 13.7 | 10.1 | 1.9 | 14.2 | 87.9 | 4.47 | 1.24 | |
| LLaMa-3-8B (Binary Cross Entropy)Similarity Refiner Loss=Binary Cross Entropy, Semantic Anchor=WithoutObj2026.03 | 14.3 | 2.2 | 12.7 | 9.4 | 1.8 | 13.7 | 87.9 | 4.49 | 1.22 | |
| LLaMa-3-8B (Naive Baseline)Similarity Refiner Loss=Naive Baseline, Semantic Anchor=WithoutObj2026.03 | 13.7 | 1.8 | 12.4 | 10.8 | 1.9 | 13 | 86.6 | 3.84 | 1.04 | |
| LLaMa-3-8B (Focal)Similarity Refiner Loss=Focal, Semantic Anchor=WithObj2026.03 | 12.6 | 2.2 | 11.6 | 6.8 | 1.5 | 12.5 | 88.1 | 4.56 | 1.33 | |
| SENSEDecoder Model=LLaMa-3-8BWith Obj2026.05 | 12.6 | 2.2 | 11.6 | 6.8 | 1.5 | 12.5 | 88.1 | 4.56 | 1.33 | |
| LLaMa-3-8B (Binary Cross Entropy)Similarity Refiner Loss=Binary Cross Entropy, Semantic Anchor=WithObj2026.03 | 12.4 | 1.8 | 11.3 | 6.6 | 1.3 | 12.1 | 88 | 4.54 | 1.3 | |
| LLaMa-3-8B (Contrastive Multi Label)Similarity Refiner Loss=Contrastive Multi Label, Semantic Anchor=WithObj2026.03 | 12.3 | 1.8 | 11.4 | 6.8 | 1.3 | 12.2 | 88 | 4.5 | 1.31 | |
| LLaMa-3-8B (Naive Baseline)Similarity Refiner Loss=Naive Baseline, Semantic Anchor=WithObj2026.03 | 10.5 | 1.2 | 9.9 | 5.3 | 1 | 10.6 | 87.7 | 4.24 | 1.22 | |
| Qwen2.5-7B (Naive Baseline)Similarity Refiner Loss=Naive Baseline, Semantic Anchor=WithoutObj2026.03 | 10 | 1.4 | 9 | 6.5 | 1.3 | 6 | 85.1 | 2.56 | 1.04 |