88 Quantifying and Learning Linear Symmetry-Based Disentanglement (LSBD) Table 4.8: Scores for the ModelNet40 dataset. MODEL BETA ↑ FACTOR ↑ SAP ↑ DCI ↑ MIG↑ MOD↑ DLSBD↓ VAE .995±.004 .838±.030 .013±.002 .013±.002 .009±.002 .415±.058 .393±.110 β-VAE .995±.005 .857±.045 .012±.003 .015±.003 .009±.002 .447±.067 .285±.045 CC-VAE .997±.003 .818±.093 .011±.003 .017±.004 .011±.003 .567±.063 .281±.191 FACTOR .996±.004 .856±.052 .012±.002 .014±.003 .010±.003 .444±.077 .388±.096 DIP-I .988±.009 .783±.070 .012±.002 .013±.002 .008±.001 .343±.082 .416±.142 DIP-II .994±.006 .832±.042 .013±.003 .014±.003 .011±.002 .433±.080 .379±.130 ADAGVAE .996±.006 .775±.079 .010±.006 .014±.006 .013±.004 .421±.092 .476±.218 ADAMLVAE .996±.006 .784±.055 .012±.006 .014±.005 .014±.004 .445±.040 .580±.141 QUESSARD .907±.192 .727±.384 .010±.005 .015±.007 .009±.004 .563±.108 .134±.294 LSBD-VAE .990±.009 .863±.038 .011±.003 .015±.003 .014±.003 .538±.103 .731±.068 /0 LSBD-VAE 1.000±.000 .990±.004 .012±.005 .052±.009 .020±.006 .947±.007 .041±.007 /FULL Table 4.9: Scores for COIL-100 dataset. MODEL BETA ↑ FACTOR ↑ SAP ↑ DCI ↑ MIG↑ MOD↑ DLSBD↓ VAE 1.000±.000 .674±.049 .014±.003 .016±.003 .011±.002 .986±.001 .463±.030 β-VAE 1.000±.001 .740±.024 .015±.004 .014±.004 .013±.003 .982±.001 .579±.095 CC-VAE .999±.003 .723±.026 .013±.005 .014±.003 .013±.004 .985±.001 .406±.057 FACTOR 1.000±.001 .684±.041 .014±.002 .012±.002 .013±.004 .984±.001 .490±.024 DIP-I .999±.002 .631±.025 .013±.004 .012±.002 .010±.002 .986±.001 .525±.109 DIP-II 1.000±.001 .643±.043 .013±.003 .014±.002 .011±.002 .985±.001 .568±.079 ADAGVAE 1.000±.000 .672±.021 .015±.007 .016±.005 .014±.006 .984±.001 .431±.049 ADAMLVAE 1.000±.000 .688±.027 .011±.003 .015±.006 .018±.009 .984±.002 .400±.076 QUESSARD 1.000±.000 .780±.044 .014±.004 .014±.002 .011±.003 .973±.004 .396±.055 LSBD-VAE 1.000±.001 .739±.047 .014±.003 .014±.001 .011±.001 .982±.004 .515±.099 /0 LSBD-VAE 1.000±.000 .655±.028 .015±.004 .029±.003 .013±.003 .802±.056 .112±.026 /FULL
RkJQdWJsaXNoZXIy MjY0ODMw