Group Activity Recognition on NBA (test)
83.3MCASOGAR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SOGARExtractor=ViT-Base, Input Type=whole image, Fine-tuning=true2026.04 | 83.3 | — | — | |
| Flaming-NetExtractor=Inception-v3, Input Type=whole image, Fine-tuning=true2026.04 | 79.1 | — | — | |
| MP-GCNExtractor=YOLOV8x, Input Type=person bounding boxes + whole image, Fine-tuning=true2026.04 | 78.7 | — | — | |
| DFWSGARExtractor=ResNet-18, Input Type=whole image, Fine-tuning=true2026.04 | 75.8 | — | — | |
| Group-DINOmicsExtractor=DINOv3, Input Type=person bounding boxes + whole image, Fine-tuning=true2026.04 | 73 | — | — | |
| KRGFormerExtractor=Inception-v3, Input Type=person bounding boxes + whole image, Fine-tuning=true2026.04 | 72.4 | — | — | |
| LiGARExtractor=ResNet-18, Input Type=whole image, Fine-tuning=true2026.04 | 62.7 | — | — | |
| Dual-AIExtractor=Inception-v3, Input Type=person bounding boxes + whole image, Fine-tuning=true2026.04 | 58.1 | — | — | |
| SAMExtractor=ResNet-18, Input Type=person bounding boxes + whole image, Fine-tuning=true2026.04 | 54.3 | — | — | |
| DFWSGAR# Params=17.5M, FLOPs=313G, Backbone=ResNet-182022.04 | 0.758 | 0.712 | — | |
| DFWSGAR w/o motion# Params=17.3M, FLOPs=311G, Backbone=ResNet-182022.04 | 0.736 | 0.69 | — | |
| TSM# Params=11.2M, FLOPs=303G, Backbone=ResNet-182022.04 | 0.666 | 0.603 | — | |
| VideoSwin# Params=27.9M, FLOPs=478G, Backbone=VideoSwin-T2022.04 | 0.643 | 0.606 | — | |
| DIN# Params=26.0M, FLOPs=304G, Backbone=ResNet-182022.04 | 0.616 | 0.56 | — | |
| ARG# Params=49.5M, FLOPs=307G, Backbone=ResNet-182022.04 | 0.59 | 0.568 | — | |
| SACRF# Params=53.7M, FLOPs=339G, Backbone=ResNet-182022.04 | 0.563 | 0.528 | — | |
| SAMsource=reproduced by current paper authors, # Params=25.5M, FLOPs=304G, Backbone=ResNet-182022.04 | 0.543 | 0.515 | — | |
| SAMsource=copied directly from original paper, Backbone=ResNet-182022.04 | 0.491 | 0.475 | — | |
| AT# Params=29.6M, FLOPs=305G, Backbone=ResNet-182022.04 | 0.471 | 0.415 | — | |
| DFWSGARExtractor=ResNet-18, Input Source=Whole image2026.04 | — | 71.2 | — | |
| Dual-AIExtractor=Inception-v3, Input Source=Person bounding boxes + whole image2026.04 | — | 50.2 | — | |
| Flaming-NetExtractor=Inception-v3, Input Source=Whole image2026.04 | — | 76 | — | |
| Group-DINOmicsExtractor=DINOv3, Input Source=Person bounding boxes + whole image2026.04 | — | 68.2 | — | |
| KRGFormerExtractor=Inception-v3, Input Source=Person bounding boxes + whole image2026.04 | — | 67.1 | — | |
| LiGARExtractor=ResNet-18, Input Source=Whole image2026.04 | — | 57.1 | — | |
| MP-GCNExtractor=YOLOv8x, Input Source=Person bounding boxes + whole image2026.04 | — | 74.6 | — | |
| SAMExtractor=ResNet-18, Input Source=Person bounding boxes + whole image2026.04 | — | 51.5 | — | |
| SOGARExtractor=ResNet-18, Input Source=Whole image2026.04 | — | 73.5 | — |