Action Recognition On Epic Kitchens 100

Métriques

Action@1

GFLOPs

Noun@1

Verb@1

Résultats

Résultats de performance de divers modèles sur ce benchmark

Nom du modèle	Action@1	GFLOPs	Noun@1	Verb@1	Paper Title	Repository
MoViNet-A5	44.5	74.9x1	55.1	69.1	MoViNets: Mobile Video Networks for Efficient Video Recognition
Avion (ViT-L)	54.4	-	65.4	73.0	Training a Large Video Model on a Single Machine in a Day
MeMViT-24	48.4	-	60.3	71.4	MeMViT: Memory-Augmented Multiscale Vision Transformer for Efficient Long-Term Video Recognition
SlowFast	36.81	-	-	-	Rescaling Egocentric Vision
MoViNet-A2	41.2	7.59x1	52.3	67.1	MoViNets: Mobile Video Networks for Efficient Video Recognition
TSN	33.57	-	-	-	Rescaling Egocentric Vision
GSF	44.48	-	53.18	69.06	Gate-Shift-Fuse for Video Action Recognition
ViViT-L/16x2 Fact. encoder	44.0	-	56.8	66.4	ViViT: A Video Vision Transformer
TAdaConvNeXtV2-S	48.9	-	60.2	71.0	Temporally-Adaptive Models for Efficient Video Understanding
ORViT Mformer-L (ORViT blocks)	45.7	-	58.7	68.4	Object-Region Video Transformers
CAST-B/16	49.3	-	60.9	72.5	CAST: Cross-Attention in Space and Time for Video Action Recognition
Mformer-HR	44.5	-	58.5	67.0	Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers
TempAgg	45.26	-	53.35	66	Technical Report: Temporal Aggregate Representations
LaViLa (TimeSformer-L)	51	-	62.9	72	Learning Video Representations from Large Language Models
MMT	47.8	-	61.0	70.1	Multiscale Multimodal Transformer for Multimodal Action Recognition	-
MBT	43.4	-	58	64.8	Attention Bottlenecks for Multimodal Fusion
Mformer-L	44.1	-	57.6	67.1	Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers
Mformer	43.1	-	56.5	66.7	Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers
M&M (WTS 60M)	53.6	-	66.3	72.0	M&M Mix: A Multimodal Multiview Transformer Ensemble	-
OMNIVORE (Swin-B, finetuned)	49.9	-	61.7	69.5	Omnivore: A Single Model for Many Visual Modalities

0 of 30 row(s) selected.