Zero Shot Action Recognition On Kinetics

Métriques

Top-1 Accuracy

Top-5 Accuracy

Résultats

Résultats de performance de divers modèles sur ce benchmark

Nom du modèle	Top-1 Accuracy	Top-5 Accuracy	Paper Title	Repository
OST	75.1	94.6	OST: Refining Text Knowledge with Optimal Spatio-Temporal Descriptor for General Video Recognition
DEM	23.6	49.5	Learning a Deep Embedding Model for Zero-Shot Learning
ER-ZSAR (ST+Obj)	42.1	73.1	Elaborative Rehearsal for Zero-shot Action Recognition
IMP-MoE-L	76.8	-	Alternating Gradient Descent and Mixture-of-Experts for Integrated Multimodal Perception	-
MAXI	71.6	-	MAtch, eXpand and Improve: Unsupervised Finetuning for Zero-Shot Action Recognition with Language Knowledge
BIKE	68.5	91.1	Bidirectional Cross-Modal Knowledge Exploration for Video Recognition with Pre-trained Vision-Language Models
ALE	23.4	50.3	Label-Embedding for Image Classification
Text4Vis	68.9	90.3	Revisiting Classifier: Transferring Vision-Language Models for Video Recognition
VideoCoCa	70.1	88.9	VideoCoCa: Video-Text Modeling with Zero-Shot Transfer from Contrastive Captioners	-
X-CLIP	65.2	86.1	Expanding Language-Image Pretrained Models for General Video Recognition
JigsawNet	45.9	78.8	Rethinking Zero-shot Action Recognition: Learning from Latent Atomic Actions
LoCATe-GAT	58.7	-	LoCATe-GAT: Modeling Multi-Scale Local Context and Action Relationships for Zero-Shot Action Recognition
ER-ZSAR (ST)	37.1	69.3	Elaborative Rehearsal for Zero-shot Action Recognition
GCN	22.3	49.7	All About Knowledge Graphs for Actions	-
LanguageBind	64.1	85.7	LanguageBind: Extending Video-Language Pretraining to N-modality by Language-based Semantic Alignment
SJE(Word Embedding)	22.3	48.2	Evaluation of Output Embeddings for Fine-Grained Image Classification
DEVISE	23.8	51.0	DeViSE: A Deep Visual-Semantic Embedding Model	-
OTI（ViT-L/14）	70.6	-	Orthogonal Temporal Interpolation for Zero-Shot Video Recognition
TC-CLIP	78.1	95.7	Leveraging Temporal Contextualization for Video Action Recognition
ESZSL	22.9	48.3	An embarrassingly simple approach to zero-shot learning

0 of 20 row(s) selected.