List your company
ResearchFormerly WizWand

Facial Expression Recognition on Ava265-HMC

53.42UARSynHMC
13.005623.497833.9944.4822Jul 5, 2026
Updated 3mo ago

Evaluation results

MethodLinks
SynHMCPre-training Dataset=SynHMC-7K, Reconstruction Method=SynHMC (Ours), Evaluation Protocol=Fine-tune, Backbone=Swin-T2026.0753.4254.4
FFHQ-UVPre-training Dataset=SynHMC-7K, Reconstruction Method=FFHQ-UV [3], Evaluation Protocol=Fine-tune, Backbone=Swin-T2026.0751.7652.9
FDSPre-training Dataset=SynHMC-7K, Reconstruction Method=FDS [7], Evaluation Protocol=Fine-tune, Backbone=Swin-T2026.0750.7352.05
AffectNet-7KPre-training Dataset=AffectNet-7K [36], Reconstruction Method=None (Frontal-view images), Evaluation Protocol=Fine-tune, Backbone=Swin-T2026.0750.2751.75
EMOCAPre-training Dataset=SynHMC-7K, Reconstruction Method=EMOCA [10], Evaluation Protocol=Fine-tune, Backbone=Swin-T2026.0749.4150.89
DECAPre-training Dataset=SynHMC-7K, Reconstruction Method=DECA [15], Evaluation Protocol=Fine-tune, Backbone=Swin-T2026.0746.8248.25
ImageNet Pre-trainedPre-training Dataset=ImageNet [11], Reconstruction Method=None (Frontal-view images), Evaluation Protocol=Fine-tune, Backbone=Swin-T2026.0746.447.96
SynHMCPre-training Dataset=SynHMC-7K, Reconstruction Method=SynHMC (Ours), Evaluation Protocol=K-NN, Backbone=Swin-T2026.0740.5342.03
SynHMCPre-training Dataset=SynHMC-7K, Reconstruction Method=SynHMC (Ours), Evaluation Protocol=Linear Probe, Backbone=Swin-T2026.0738.2741.04
FDSPre-training Dataset=SynHMC-7K, Reconstruction Method=FDS [7], Evaluation Protocol=K-NN, Backbone=Swin-T2026.0737.6639.13
FFHQ-UVPre-training Dataset=SynHMC-7K, Reconstruction Method=FFHQ-UV [3], Evaluation Protocol=K-NN, Backbone=Swin-T2026.0736.5838.33
FDSPre-training Dataset=SynHMC-7K, Reconstruction Method=FDS [7], Evaluation Protocol=Linear Probe, Backbone=Swin-T2026.0735.6838.42
FFHQ-UVPre-training Dataset=SynHMC-7K, Reconstruction Method=FFHQ-UV [3], Evaluation Protocol=Linear Probe, Backbone=Swin-T2026.0734.2337.33
DECAPre-training Dataset=SynHMC-7K, Reconstruction Method=DECA [15], Evaluation Protocol=K-NN, Backbone=Swin-T2026.0733.7335.42
AffectNet-7KPre-training Dataset=AffectNet-7K [36], Reconstruction Method=None (Frontal-view images), Evaluation Protocol=K-NN, Backbone=Swin-T2026.0733.6534.43
EMOCAPre-training Dataset=SynHMC-7K, Reconstruction Method=EMOCA [10], Evaluation Protocol=K-NN, Backbone=Swin-T2026.0733.2734.73
EMOCAPre-training Dataset=SynHMC-7K, Reconstruction Method=EMOCA [10], Evaluation Protocol=Linear Probe, Backbone=Swin-T2026.0732.0435.25
AffectNet-7KPre-training Dataset=AffectNet-7K [36], Reconstruction Method=None (Frontal-view images), Evaluation Protocol=Linear Probe, Backbone=Swin-T2026.0731.7434.04
DECAPre-training Dataset=SynHMC-7K, Reconstruction Method=DECA [15], Evaluation Protocol=Linear Probe, Backbone=Swin-T2026.0729.7832.84
SynHMCPre-training Dataset=SynHMC-7K, Reconstruction Method=SynHMC (Ours), Evaluation Protocol=Direct Test, Backbone=Swin-T2026.0727.9327.79
FFHQ-UVPre-training Dataset=SynHMC-7K, Reconstruction Method=FFHQ-UV [3], Evaluation Protocol=Direct Test, Backbone=Swin-T2026.0725.7826.49
MEADPre-training Dataset=MEAD [55], Reconstruction Method=None (Frontal-view images), Evaluation Protocol=Linear Probe, Backbone=Swin-T2026.0723.6826.31
MEADPre-training Dataset=MEAD [55], Reconstruction Method=None (Frontal-view images), Evaluation Protocol=K-NN, Backbone=Swin-T2026.0723.4925.4
MEADPre-training Dataset=MEAD [55], Reconstruction Method=None (Frontal-view images), Evaluation Protocol=Fine-tune, Backbone=Swin-T2026.0722.4525.97
FDSPre-training Dataset=SynHMC-7K, Reconstruction Method=FDS [7], Evaluation Protocol=Direct Test, Backbone=Swin-T2026.0721.4519.62
DECAPre-training Dataset=SynHMC-7K, Reconstruction Method=DECA [15], Evaluation Protocol=Direct Test, Backbone=Swin-T2026.0721.3520.02
EMOCAPre-training Dataset=SynHMC-7K, Reconstruction Method=EMOCA [10], Evaluation Protocol=Direct Test, Backbone=Swin-T2026.0719.3818.75
AffectNet-7KPre-training Dataset=AffectNet-7K [36], Reconstruction Method=None (Frontal-view images), Evaluation Protocol=Direct Test, Backbone=Swin-T2026.0719.1318.2
MEADPre-training Dataset=MEAD [55], Reconstruction Method=None (Frontal-view images), Evaluation Protocol=Direct Test, Backbone=Swin-T2026.0715.7816.27
Randomly Initialized FERPre-training Dataset=None, Reconstruction Method=None (Frontal-view images), Evaluation Protocol=Fine-tune, Backbone=Swin-T2026.0714.5617.82