Video Question Answering on LongVideoBench KFS-Bench (full set)
65.5AccuracyASCS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ASCSMLLM=Qwen2.5-VL-32B, Frames=64, Sampling Method=ASCS2025.12 | 65.5 | 0.269 | |
| ASCSMLLM=InternVL3-8B, Frames=64, Sampling Method=ASCS2025.12 | 65.4 | 0.268 | |
| ITSMLLM=InternVL3-8B, Frames=64, Sampling Method=ITS2025.12 | 65.1 | 0.256 | |
| AKSMLLM=InternVL3-8B, Frames=64, Sampling Method=AKS2025.12 | 64.9 | 0.235 | |
| ASCSMLLM=Qwen2.5-VL-7B, Frames=64, Sampling Method=ASCS2025.12 | 64.1 | 0.269 | |
| ASCSMLLM=Qwen2.5-VL-7B, Frames=32, Sampling Method=ASCS2025.12 | 63.6 | 0.199 | |
| ITSMLLM=Qwen2.5-VL-32B, Frames=64, Sampling Method=ITS2025.12 | 63.6 | 0.256 | |
| AKSMLLM=Qwen2.5-VL-32B, Frames=64, Sampling Method=AKS2025.12 | 63.5 | 0.235 | |
| ITSMLLM=Qwen2.5-VL-7B, Frames=64, Sampling Method=ITS2025.12 | 63.4 | 0.256 | |
| AKSMLLM=Qwen2.5-VL-7B, Frames=64, Sampling Method=AKS2025.12 | 63.2 | 0.235 | |
| K-meansMLLM=InternVL3-8B, Frames=64, Sampling Method=K-means2025.12 | 62.7 | 0.157 | |
| ITSMLLM=Qwen2.5-VL-7B, Frames=32, Sampling Method=ITS2025.12 | 61.9 | 0.202 | |
| K-meansMLLM=Qwen2.5-VL-32B, Frames=64, Sampling Method=K-means2025.12 | 61.7 | 0.157 | |
| AKSMLLM=Qwen2.5-VL-7B, Frames=32, Sampling Method=AKS2025.12 | 61.5 | 0.188 | |
| UniformMLLM=InternVL3-8B, Frames=64, Sampling Method=Uniform2025.12 | 60.8 | 0.143 | |
| UniformMLLM=Qwen2.5-VL-32B, Frames=64, Sampling Method=Uniform2025.12 | 60.2 | 0.143 | |
| K-meansMLLM=Qwen2.5-VL-7B, Frames=32, Sampling Method=K-means2025.12 | 59.8 | 0.11 | |
| UniformMLLM=Qwen2.5-VL-7B, Frames=64, Sampling Method=Uniform2025.12 | 59.5 | 0.143 | |
| K-meansMLLM=Qwen2.5-VL-7B, Frames=64, Sampling Method=K-means2025.12 | 59.5 | 0.157 | |
| UniformMLLM=Qwen2.5-VL-7B, Frames=32, Sampling Method=Uniform2025.12 | 57.2 | 0.099 |