Multimodal Understanding on SEED-I Image
0.822AccuracyDOPCD
Evaluation Results
| Method | Links | |
|---|---|---|
| DOPCDSparsity Op.=true, Sparsity Tok.=true, Vision-side Token=32, Vision-side TFLOPs (Rel.)=0.42 (5.4%)2026.02 | 0.822 | |
| FLARE-X 8B# Vis tok.=14002025.04 | 0.787 | |
| InternVL2# Parameters=40B2024.12 | 0.782 | |
| GPT-4.1LLM=–, Tokens=–, Zero-shot=true2025.07 | 0.78 | |
| FLARE-L 8B# Vis tok.=6302025.04 | 0.776 | |
| Qwen2.5-VLLLM=Qwen2.5-7B, Tokens=–, Zero-shot=true2025.07 | 0.77 | |
| Qwen2.5VL 7B# Vis tok.=14002025.04 | 0.77 | |
| InternVL3.5LLM=Qwen3-14B, Tokens=3.96B, Zero-shot=true2025.07 | 0.768 | |
| NVILA# Parameters=15B2024.12 | 0.766 | |
| InternVL3.5-REGATELLM=Qwen3-14B, Tokens=2.32B (↓ 41.41%), Zero-shot=true2025.07 | 0.766 | |
| NVILA# Parameters=8B2024.12 | 0.765 | |
| GPT-4oLLM=–, Tokens=–, Zero-shot=true2025.07 | 0.764 | |
| NVILA-Lite# Parameters=8B2024.12 | 0.763 | |
| FLARE-X 3B# Vis tok.=14002025.04 | 0.763 | |
| Eagle 8B# Vis tok.=10242025.04 | 0.763 | |
| GPT-4o2024.12 | 0.762 | |
| InternVL2# Parameters=8B2024.12 | 0.762 | |
| Qwen2-VL# Parameters=8B2024.12 | 0.76 | |
| Gemini-1.5-ProLLM=–, Tokens=–, Zero-shot=true2025.07 | 0.76 | |
| LLaVA-NeXT# Parameters=34B2024.12 | 0.759 | |
| NVILA-Lite# Parameters=15B2024.12 | 0.756 | |
| LLaVA-OneVision# Parameters=8B2024.12 | 0.754 | |
| LLaVA-OneVision# Parameters=72B2024.12 | 0.754 | |
| LLaVA-OneVisionLLM=Qwen2-7B, Tokens=–, Zero-shot=true2025.07 | 0.754 | |
| Cambrian-1# Parameters=34B2024.12 | 0.753 | |
| Gemini-1.5-FlashLLM=–, Tokens=–, Zero-shot=true2025.07 | 0.75 | |
| Florence-VL# Parameters=8B2024.12 | 0.749 | |
| Florence-VL 8B# Vis tok.=5762025.04 | 0.749 | |
| Molmo 7B# Vis tok.=12002025.04 | 0.749 | |
| LLaVA-OneVision 7B# Vis tok.=24002025.04 | 0.748 | |
| Cambrian-1# Parameters=8B2024.12 | 0.747 | |
| Cambrain 8B# Vis tok.=5762025.04 | 0.747 | |
| Cambrian-1# Parameters=13B2024.12 | 0.744 | |
| Claude3.7-SonnetLLM=–, Tokens=–, Zero-shot=true2025.07 | 0.743 | |
| FLARE-L 3B# Vis tok.=6302025.04 | 0.742 | |
| Qwen2.5VL 3B# Vis tok.=14002025.04 | 0.74 | |
| VILA-1.5# Parameters=8B2024.12 | 0.738 | |
| VILA-1.5# Parameters=13B2024.12 | 0.726 | |
| LLaVA-NextLLM=Mistral-7B, Tokens=–, Zero-shot=true2025.07 | 0.724 | |
| DeepSeek-VL2-Tiny# Vis tok.=15002025.04 | 0.723 | |
| MiniCPM-Llama3-V-2.5 8B# Vis tok.=4002025.04 | 0.723 | |
| Florence-VL 3B# Vis tok.=5762025.04 | 0.706 | |
| VideoLLaMA2LLM=Qwen2-7B, Tokens=83.82M, Zero-shot=true2025.07 | 0.704 | |
| LLaVA-1.6LLM=Vicuna-7B, Tokens=–, Zero-shot=true2025.07 | 0.702 | |
| VideoLLaMA2-REGATELLM=Qwen2-7B, Tokens=49.27M (↓ 41.22%), Zero-shot=true2025.07 | 0.7 | |
| Phi 3.5-Vision# Vis tok.=21002025.04 | 0.699 | |
| LLaVA-1.5+CoSLanguage=Vicuna-13B2024.03 | 0.696 | |
| VILA-1.5# Parameters=40B2024.12 | 0.691 | |
| LLaVA-1.5Language=Vicuna-13B2024.03 | 0.682 | |
| LLaVA-1.5# Parameters=13B2024.12 | 0.682 | |
| LLaVA-1.5+CoSLanguage=Vicuna-7B2024.03 | 0.671 | |
| MiniCPM-V-2.0 3B# Vis tok.=4002025.04 | 0.671 | |
| LLaVA-1.5-7BSparsity Op.=false, Sparsity Tok.=false, Vision-side Token=576, Vision-side TFLOPs (Rel.)=7.65 (100.0%)2026.02 | 0.662 | |
| LLaVA-1.5Language=Vicuna-7B2024.03 | 0.661 | |
| LLaVA-1.5# Parameters=7B2024.12 | 0.661 | |
| LLaVA-1.5LLM=Vicuna-7B, Tokens=–, Zero-shot=true2025.07 | 0.661 | |
| PDropSparsity Op.=false, Sparsity Tok.=true, Vision-side Token=270, Vision-side TFLOPs (Rel.)=3.54 (46.3%)2026.02 | 0.658 | |
| Qwen-VL-ChatLanguage=Qwen-7B2024.03 | 0.654 | |
| Dynamic-LLaVASparsity Op.=false, Sparsity Tok.=true, Vision-side Token=115, Vision-side TFLOPs (Rel.)=1.50 (19.6%)2026.02 | 0.646 | |
| YOPOSparsity Op.=true, Sparsity Tok.=false, Vision-side Token=70, Vision-side TFLOPs (Rel.)=0.92 (12.0%)2026.02 | 0.646 | |
| ViCASparsity Op.=false, Sparsity Tok.=true, Vision-side Token=24, Vision-side TFLOPs (Rel.)=0.31 (4.1%)2026.02 | 0.632 | |
| VILA1.5LLM=Llama-2-13B, Tokens=–, Zero-shot=true2025.07 | 0.628 | |
| Qwen-VLLanguage=Qwen-7B2024.03 | 0.623 | |
| TRIMSparsity Op.=false, Sparsity Tok.=true, Vision-side Token=29, Vision-side TFLOPs (Rel.)=0.38 (4.9%)2026.02 | 0.618 | |
| ViCA+PDrop†Sparsity Op.=true, Sparsity Tok.=true, Vision-side Token=12, Vision-side TFLOPs (Rel.)=0.16 (2.0%)2026.02 | 0.608 | |
| InstructBLIPLanguage=Vicuna-13B2024.03 | 0.588 | |
| DivPruneSparsity Op.=false, Sparsity Tok.=true, Vision-side Token=32, Vision-side TFLOPs (Rel.)=0.42 (5.4%)2026.02 | 0.587 | |
| Qwen-VL-ChatLLM=Qwen-7B, Tokens=–, Zero-shot=true2025.07 | 0.582 | |
| TwigVLMSparsity Op.=false, Sparsity Tok.=true, Vision-side Token=64, Vision-side TFLOPs (Rel.)=0.83 (10.9%)2026.02 | 0.569 | |
| DOPvSparsity Op.=true, Sparsity Tok.=true, Vision-side Token=32, Vision-side TFLOPs (Rel.)=0.42 (5.4%)2026.02 | 0.567 | |
| VisPrunerSparsity Op.=false, Sparsity Tok.=true, Vision-side Token=32, Vision-side TFLOPs (Rel.)=0.42 (5.4%)2026.02 | 0.543 | |
| LLaVA-PruMergeSparsity Op.=false, Sparsity Tok.=true, Vision-side Token=32, Vision-side TFLOPs (Rel.)=0.42 (5.4%)2026.02 | 0.507 | |
| VideoChat2-REGATELLM=Mistral-7B, Tokens=2.22B (↓ 43.51%), Zero-shot=true2025.07 | 0.472 | |
| InstructBLIPLLM=Vicuna-7B, Tokens=–, Zero-shot=true2025.07 | 0.464 | |
| VideoChat2LLM=Mistral-7B, Tokens=3.93B, Zero-shot=true2025.07 | 0.459 |