Skip to main content
. 2021 Dec 14;21(13):6. doi: 10.1167/jov.21.13.6

Table 2.

Error rates for different OSCAR v2 datasets and all model architectures.

OS-MNIST
Variant B B-F B-K B-D BT BL BLT GLM
Mono Centered .432±.002 .355±.001 .305±.001 .160±.001 .199±.001 .180±.001 .173±.001 .500±.000
Random .513±.002 .446±.002 .429±.001 .338±.001 .315±.002 .252±.000 .245±.002 .891±.000
Stereo Centered .209±.001 .165±.000 .139±.001 .087±.000 .102±.001 .086±.000 .085±.001 .325±.000
Random .284±.001 .237±.001 .223±.001 .192±.000 .175±.000 .137±.000 .138±.000 .887±.000
OS-fMNIST
Variant B B-F B-K B-D BT BL BLT GLM
Mono Centered .424±.001 .351±.001 .292±.001 .214±.000 .250±.000 .242±.001 .234±.000 .407±.000
Random .617±.001 .537±.001 .514±.002 .408±.000 .412±.001 .362±.001 .352±.002 .895±.000
Stereo Centered .260±.001 .225±.000 .209±.001 .177±.000 .203±.001 .190±.000 .195±.000 .303±.000
Random .363±.001 .318±.001 .322±.000 .299±.001 .295±.001 .257±.001 .261±.001 .884±.000
OS-YCB
Variant B B-F B-K B-D BT BL BLT GLM
Mono All .376±.002 .259±.001 .284±.001 .199±.001 .252±.001 .224±.001 .212±.000 .250±.000
Stereo All .166±.001 .092±.001 .105±.001 .064±.000 .090±.001 .071±.001 .069±.000 .091±.000

Standard error based on five independent training runs. Training occurred for 100 epochs, batchsize 500. Best two performances per dataset are highlighted in bold.