Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Ego-Exo4D

Benchmarks

Task NameDataset NameSOTA ResultTrend
Action RecognitionEgo-Exo4D Bike Repair
Top-1 Accuracy29.03
16
Proficiency EstimationEgo-Exo4D
Bouldering Proficiency Score65.41
16
Egocentric Text RetrievalEgo-Exo4D
Overall Top-1 Accuracy55.1
16
Cross-view Instance SegmentationEgo-Exo4D Exo-to-Ego
IoU68
15
Cross-view Instance SegmentationEgo-Exo4D Ego-to-Exo
IoU67.7
15
Relative camera pose estimationEgo-Exo4D (val)
Rotation Only AUC@538.54
13
Expert Demonstration RetrievalEgo-Exo4D 1.0 (test)
Recall@5022.5
13
Expert Commentary GenerationEgo-Exo4D 1.0 (test)
BLEU-445.8
13
Action RecognitionEgo-Exo4D Cooking (test)
Top-1 Accuracy55.74
11
Exo-to-Ego object correspondenceEgo-Exo4D Correspondences v2 (test)
IoU49.6
11
Ego-to-Exo object correspondenceEgo-Exo4D Correspondences v2 (test)
IoU46.3
11
Cross-view Object CorrespondenceEgo-Exo4D v2 (test)
Ego Query IoU42.57
11
Temporal GroundingEgo-Exo4D M views
Recall@135
10
Temporal GroundingEgo-Exo4D E views
Recall@136
10
Egocentric latent state predictionEgo-Exo4D Cooking
L2 Error (2s)0.062
7
Egocentric latent state predictionEgo-Exo4D Bike
L2 Distance (2s)0.048
7
Multi-view video understandingEgo-Exo4D Demonstrator Proficiency
Accuracy44.2
7
Closed-vocabulary procedural planningEgo-Exo4D Keystep |V|=375 (test)
Success Rate (SR)2.98
6
Keystep recognitionEgo-Exo4D
Top-1 Accuracy24.07
6
Action RecognitionEgo-Exo4D Cooking
Top-1 Accuracy55.74
5
Temporal GroundingEgo-Exo4D D views
Recall@128
5
Exo-to-Ego Video GenerationEgo-Exo4D Cooking
PSNR14.3897
5
Exo-to-Ego Video GenerationEgo-Exo4D Bike
PSNR15.6301
5
Exo-to-Ego Video GenerationEgo-Exo4D Health
PSNR16.7139
5
Instructional Streaming Video GenerationEgo-Exo4D KeyStep (val)
Average Score0.361
5
Showing 25 of 37 rows