Group Activity Detection on Cafe (Split-by-View)
19.67Group mAP@1.0VLCMTN
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VLCMTNBackbone=SAM, Full FT=false, Trainable Param(M)=∼ 24*2025.08 | 19.67 | 41.06 | 68.29 | |
| Practical GADBackbone=ResNet-18, Full FT=true, Trainable Param(M)=22.772025.08 | 18.84 | 37.53 | 67.64 | |
| ProGraD (Decoder + Prompts)Backbone=DINOv2, Full FT=false, Trainable Param(M)=10.742025.08 | 16.85 | 43.42 | 70.42 | |
| ProGraD (Decoder + Adapter)Backbone=DINOv2, Full FT=false, Trainable Param(M)=11.872025.08 | 16.17 | 42.67 | 68.56 | |
| ProGraD (Decoder Only)Backbone=DINOv2, Full FT=false, Trainable Param(M)=10.682025.08 | 15.88 | 41.58 | 68.55 | |
| JRDB-baseBackbone=I3D, Full FT=true, Trainable Param(M)=∼16*2025.08 | 15.43 | 34.81 | 60.43 | |
| JointBackbone=I3D, Full FT=true, Trainable Param(M)=∼17*2025.08 | 14.05 | 36.08 | 60.09 | |
| ARGBackbone=Inception-v3, Full FT=true, Trainable Param(M)=∼52*2025.08 | 11.03 | 34.5 | 56.61 | |
| HGCBackbone=I3D, Full FT=true, Trainable Param(M)=∼20*2025.08 | 6.55 | 26.29 | 56.84 |