Spatio-temporal Action Localization on AVA-Kinetics v1.0 (val)
40.52mAP@0.5 (AVA)ensemble
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ensemblepretrain=mixed, head=mixed2021.06 | 40.52 | 37.04 | 40.97 | |
| ir-CSN-152pretrain=K700, input=64x1, head=T-only, memory_bank=true2021.06 | 37.95 | 35.26 | 38.43 | |
| SlowFast-152pretrain=K700, input=16x4+64x1, head=T-only, memory_bank=true2021.06 | 37.68 | 35.54 | 38.12 | |
| ir-CSN-152pretrain=K700, input=32x2, head=T-only, memory_bank=true2021.06 | 37.32 | 35.01 | 37.78 | |
| ir-CSN-152pretrain=K400+IG65M, input=32x2, head=T-only, memory_bank=true2021.06 | 36.99 | 33.44 | 37.34 | |
| SlowFast-152pretrain=K700, input=8x8+32x2, head=T-only, memory_bank=true2021.06 | 36.88 | 33.44 | 37.35 | |
| ir-CSN-152pretrain=K700, input=32x2, head=S-only, memory_bank=false2021.06 | 36.69 | 34.12 | 37.16 | |
| ir-CSN-152pretrain=K700, input=32x2, head=T-only, memory_bank=false2021.06 | 36.41 | 34.8 | 36.98 | |
| SlowFast-152pretrain=K700, input=8x8+32x2, head=T-only, memory_bank=false2021.06 | 36.34 | 33.16 | 36.88 | |
| ir-CSN-152pretrain=K400+IG65M, input=32x2, head=T-only, memory_bank=false2021.06 | 36.29 | 32.81 | 36.58 | |
| SlowFast-152pretrain=K700, input=8x8+32x2, head=S-only, memory_bank=false2021.06 | 36.16 | 33.82 | 36.92 | |
| ir-CSN-152pretrain=K400+IG65M, input=32x2, head=S-only, memory_bank=false2021.06 | 36.07 | 32.62 | 36.58 | |
| ir-CSN-152pretrain=K700, input=32x2, head=Linear, memory_bank=false2021.06 | 35.67 | 33.2 | 36.22 | |
| ir-CSN-152pretrain=K400+IG65M, input=32x2, head=Linear, memory_bank=false2021.06 | 35.38 | 32.58 | 36.2 | |
| SlowFast-152pretrain=K700, input=8x8+32x2, head=Linear, memory_bank=false2021.06 | 35.29 | 32.88 | 36.26 |