Action Recognition
Action recognition is a task in the field of computer vision that aims to identify and classify human behaviors through videos or images. Its goal is to categorize the actions being performed in videos or images into predefined action categories, thereby achieving accurate action detection and understanding. This task holds significant value for applications such as video surveillance, human-computer interaction, and sports analysis. However, the challenge of building large-scale video datasets has led to most existing action recognition benchmarks being relatively small, typically containing only around 10k videos.
Something-Something V2
MSNet-R50En (8+16 ensemble, ImageNet pretrained)
UCF101
ResNet50
HMDB-51
VideoMAE V2-g
Something-Something V1
InternVideo
AVA v2.2
VideoMAE (K700 pretrain+finetune, ViT-L, 16x4)
EPIC-KITCHENS-100
Avion (ViT-L)
NTU RGB+D
PoseC3D (RGB + Pose)
NTU RGB+D 120
PoseC3D (RGB + Pose)
Diving-48
ActivityNet
Text4Vis (w/ ViT-L)
AVA v2.1
H2O (2 Hands and Objects)
HandFormer-B/21x8
THUMOS’14
BMN
Sports-1M
ip-CSN-152 (RGB)
HACS
UniFormerV2-L
Charades-Ego
LaViLa (Finetuned, TimeSformer-L)
Volleyball
PoseC3D (Pose Only)
BAR
HAA500
UAV-Human
PMI Sampler
Animal Kingdom
Jester (Gesture Recognition)
DirecFormer
RareAct
Real Life Violence Situations Dataset
DeVTr
ICVL-4
IRD
miniSports
UCF-101
R3D-18
Drone-Action
Mimetics
JMRN
Okutama-Action
Penn Action
SL-Animals
SEW-Resnet18 (3sets)
ActionNet-VE
Charades
EgoGesture
EPIC-KITCHENS-55
HMDB51
MSQNet
UTD-MHAD
VIRAT Ground 2.0
DVS128 Gesture
Hockey
IndustReal
KTH
CNN-GRU
MECCANO
SlowFast
MTL-AQA
C3D-AVG
N-UCLA
DVANet
NEC Drone
RoCoG-v2
Skeleton-Mimetics
THUMOS14
UAV Human
FAR
UCF 101
R2+1D-BERT
UCFSports
Win-Fail Action Understanding
2DCNN+TRN