Streaming Video Captioning on LiveSports3k (test)
43.2Winrate (%)LiveCC-7B-Base
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| LiveCC-7B-BaseParameters=7B2025.12 | 43.2 | — | — | — | — | |
| LiveCC-7B-InstructParameters=7B2025.12 | 41.5 | — | — | — | — | |
| CASA⊕ Qwen2.5-VLTraining steps=17k, Parameters=3B2025.12 | 39.4 | — | — | — | — | |
| CASA⊕ Qwen2.5-VLTraining steps=20k, Parameters=3B2025.12 | 39 | — | — | — | — | |
| CASA⊕ Qwen2.5-VLTraining steps=15k, Parameters=3B2025.12 | 36.3 | — | — | — | — | |
| CASA⊕ Qwen2.5-VLTraining steps=12k, Parameters=3B2025.12 | 34.3 | — | — | — | — | |
| CASA⊕ Qwen2.5-VLTraining steps=8k, Parameters=3B2025.12 | 33.7 | — | — | — | — | |
| Qwen2-VL-7B-LiveCCParameters=7B2025.12 | 33.7 | — | — | — | — | |
| CASA⊕ Qwen2.5-VLTraining steps=6k, Parameters=3B2025.12 | 27.6 | — | — | — | — | |
| CASA⊕ Qwen2.5-VLTraining steps=3k, Parameters=3B2025.12 | 25.4 | — | — | — | — | |
| Livecc-7B-InstructInference Mode=infinite2025.10 | — | 41.5 | 40.06 | 39.73 | — | |
| Qwen-2.5-VL-7B-InstructInference Mode=chunked, Chunk Length=100s2025.10 | — | 24.5 | 16.25 | 28.38 | 34.11 | |
| StreamingVLMInference Mode=infinite2025.10 | — | 47.33 | 45.59 | 44.21 | 56.19 |