Data Splits and Metrics for Method Benchmarking on Surgical Action Triplet Datasets

Benchmark Model Rank Results
action-triplet-recognition-on-cholect45Rendezvous#1mAP: 29.4±2.8
action-triplet-recognition-on-cholect45Attention Tripnet#2mAP: 27.2±2.7
action-triplet-recognition-on-cholect45Tripnet#3mAP: 24.4±4.7
action-triplet-recognition-on-cholect50-cross-valRendezvous#1mAP: 29.4±2.5
action-triplet-recognition-on-cholect50-cross-valAttention Tripnet#2mAP: 27.2±2.9
action-triplet-recognition-on-cholect50-cross-valTripnet#3mAP: 25.3±2.4
action-triplet-recognition-on-cholect45-cross-valRendezvous#2mAP: 29.4±2.8
action-triplet-recognition-on-cholect45-cross-valAttention Tripnet#3mAP: 27.2±2.7
action-triplet-recognition-on-cholect45-cross-valTripnet#4mAP: 24.4±4.7
action-triplet-recognition-on-cholect50Rendezvous (PyTorch)#2Mean AP: 29.5
action-triplet-recognition-on-cholect50Attention Tripnet (PyTorch)#4Mean AP: 23.3
action-triplet-recognition-on-cholect50Tripnet (PyTorch)#5Mean AP: 21.6
action-triplet-recognition-on-cholect50-challengeRendezvous (PyTorch)#6mAP: 32.8
action-triplet-recognition-on-cholect50-challengeAttention Tripnet (PyTorch)#12mAP: 27.7
action-triplet-recognition-on-cholect50-challengeTripnet (PyTorch)#13mAP: 27.4