Fine-grained Activity Recognition on Newborn Resuscitation dataset (test)
92F1 Score (Vent.)FT-C-LoRA
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| FT-C-LoRAEvaluation Protocol=LoRA-based classifier head, Backbone=LLaVA-NeXT Video, Input Frames=162026.02 | 92 | 99.6 | 79 | 94 | 91 | |
| FT-LCEvaluation Protocol=Fine-tuned classification head, Backbone=LLaVA-NeXT Video, Input Frames=162026.02 | 60 | 99 | 68 | 19 | 62 | |
| TimeSformerPre-training=SSV2, Input Frames=8, Evaluation Protocol=Fine-tuned2026.02 | 51 | 99 | 71 | 57 | 70 | |
| ZS-BEvaluation Protocol=Zero-shot, Backbone=LLaVA-NeXT Video, Input Frames=162026.02 | 46 | 99 | 50 | 19 | 54 | |
| ZSC-COEvaluation Protocol=Zero-shot, Backbone=LLaVA-NeXT Video, Input Frames=162026.02 | 45 | — | 24 | 22 | 23 | |
| ZSC-JEvaluation Protocol=Zero-shot, Backbone=LLaVA-NeXT Video, Input Frames=162026.02 | 40 | 92 | 36 | 11 | 45 |