Sor0ush/geometry-binding-repro / code /outputs /fig_paper_table_correlations.json
Sor0ush's picture
download
raw
9.02 kB
{
"summary": {
"n_models": 26,
"models": [
"Barlow Twins",
"VICReg",
"SwAV",
"DINOv2 ViT-B/14",
"DINOv2 ViT-S/14",
"DINOv2 ViT-L/14",
"MoCo v3",
"MAE ViT-L/16",
"DINO ViT-S/16",
"DINO ViT-B/16",
"DINOv2 ViT-g/14",
"MAE ViT-B/16",
"ViT-L/16",
"BEiTv2",
"ConvNeXt Large",
"ViT-B/16",
"ConvNeXt Base",
"CLIP ViT-B/16",
"CLIP ViT-L/14",
"SigLIP ViT-B/16",
"SigLIP SoViT-400M",
"BEiT",
"CLIP ViT-B/32",
"EVA-CLIP ViT-E/14",
"EVA-CLIP ViT-B/16",
"EVA-CLIP ViT-L/14"
],
"G.PR": {
"r": 0.10175706384849538,
"p": 0.6208611210490093,
"n": 26,
"R2": 0.010354500043066817
},
"G.Iso": {
"r": 0.11043394682061311,
"p": 0.5912275049495603,
"n": 26,
"R2": 0.012195656610377692
},
"L.Iso": {
"r": 0.006045326440772988,
"p": 0.9766178387496675,
"n": 26,
"R2": 3.654597177549235e-05
},
"JER": {
"r": 0.6871683052539043,
"p": 0.00010538392300148545,
"n": 26,
"R2": 0.4722002797455227
},
"disc": {
"r": 0.7242320872233026,
"p": 2.877554112384543e-05,
"n": 26,
"R2": 0.5245121161638209
},
"JER+Disc": {
"R2": 0.7829973669623611,
"LOO_R2": 0.7139457932658042,
"n": 26
},
"by_family": {
"Var-Decorr": {
"binding_mean": 0.441,
"JER_mean": 28.8,
"models": [
"Barlow Twins",
"VICReg"
]
},
"Clustering": {
"binding_mean": 0.34,
"JER_mean": 29.9,
"models": [
"SwAV"
]
},
"Self-Distill": {
"binding_mean": 0.23299999999999998,
"JER_mean": 24.349999999999998,
"models": [
"DINOv2 ViT-B/14",
"DINOv2 ViT-S/14",
"DINOv2 ViT-L/14",
"DINO ViT-S/16",
"DINO ViT-B/16",
"DINOv2 ViT-g/14"
]
},
"Contrastive": {
"binding_mean": 0.224,
"JER_mean": 16.3,
"models": [
"MoCo v3"
]
},
"Masked": {
"binding_mean": 0.17650000000000002,
"JER_mean": 23.525,
"models": [
"MAE ViT-L/16",
"MAE ViT-B/16",
"BEiTv2",
"BEiT"
]
},
"Supervised": {
"binding_mean": 0.161,
"JER_mean": 16.2,
"models": [
"ViT-L/16",
"ConvNeXt Large",
"ViT-B/16",
"ConvNeXt Base"
]
},
"Vision-Lang": {
"binding_mean": 0.12375,
"JER_mean": 19.974999999999998,
"models": [
"CLIP ViT-B/16",
"CLIP ViT-L/14",
"SigLIP ViT-B/16",
"SigLIP SoViT-400M",
"CLIP ViT-B/32",
"EVA-CLIP ViT-E/14",
"EVA-CLIP ViT-B/16",
"EVA-CLIP ViT-L/14"
]
}
},
"claim4": {
"var_decorr": {
"binding_mean": 44.1,
"JER_mean": 28.8,
"n": 2,
"rows": [
{
"model": "Barlow Twins",
"binding": 0.446,
"JER": 28.8
},
{
"model": "VICReg",
"binding": 0.436,
"JER": 28.8
}
]
},
"clip": {
"binding_mean": 13.133333333333333,
"JER_mean": 20.400000000000002,
"n": 3,
"rows": [
{
"model": "CLIP ViT-B/16",
"binding": 0.138,
"JER": 20.8
},
{
"model": "CLIP ViT-L/14",
"binding": 0.13,
"JER": 18.7
},
{
"model": "CLIP ViT-B/32",
"binding": 0.126,
"JER": 21.7
}
]
}
}
},
"rows": [
{
"model": "Barlow Twins",
"family": "Var-Decorr",
"G.PR": 0.11,
"G.Iso": 0.91,
"L.Iso": 0.82,
"JER": 28.8,
"disc": 0.75,
"binding": 0.446
},
{
"model": "VICReg",
"family": "Var-Decorr",
"G.PR": 0.06,
"G.Iso": 0.91,
"L.Iso": 0.82,
"JER": 28.8,
"disc": 0.75,
"binding": 0.436
},
{
"model": "SwAV",
"family": "Clustering",
"G.PR": 0.08,
"G.Iso": 0.89,
"L.Iso": 0.8,
"JER": 29.9,
"disc": 0.75,
"binding": 0.34
},
{
"model": "DINOv2 ViT-B/14",
"family": "Self-Distill",
"G.PR": 0.22,
"G.Iso": 0.94,
"L.Iso": 0.82,
"JER": 24.2,
"disc": 0.746,
"binding": 0.302
},
{
"model": "DINOv2 ViT-S/14",
"family": "Self-Distill",
"G.PR": 0.29,
"G.Iso": 0.94,
"L.Iso": 0.82,
"JER": 26.8,
"disc": 0.75,
"binding": 0.27
},
{
"model": "DINOv2 ViT-L/14",
"family": "Self-Distill",
"G.PR": 0.21,
"G.Iso": 0.94,
"L.Iso": 0.83,
"JER": 23.1,
"disc": 0.724,
"binding": 0.238
},
{
"model": "MoCo v3",
"family": "Contrastive",
"G.PR": 0.11,
"G.Iso": 0.9,
"L.Iso": 0.85,
"JER": 16.3,
"disc": 0.75,
"binding": 0.224
},
{
"model": "MAE ViT-L/16",
"family": "Masked",
"G.PR": 0.01,
"G.Iso": 0.71,
"L.Iso": 0.72,
"JER": 25.5,
"disc": 0.5,
"binding": 0.22
},
{
"model": "DINO ViT-S/16",
"family": "Self-Distill",
"G.PR": 0.21,
"G.Iso": 0.9,
"L.Iso": 0.81,
"JER": 26.3,
"disc": 0.686,
"binding": 0.196
},
{
"model": "DINO ViT-B/16",
"family": "Self-Distill",
"G.PR": 0.11,
"G.Iso": 0.9,
"L.Iso": 0.83,
"JER": 24.0,
"disc": 0.696,
"binding": 0.196
},
{
"model": "DINOv2 ViT-g/14",
"family": "Self-Distill",
"G.PR": 0.14,
"G.Iso": 0.93,
"L.Iso": 0.82,
"JER": 21.7,
"disc": 0.71,
"binding": 0.196
},
{
"model": "MAE ViT-B/16",
"family": "Masked",
"G.PR": 0.0,
"G.Iso": 0.33,
"L.Iso": 0.56,
"JER": 27.9,
"disc": 0.5,
"binding": 0.182
},
{
"model": "ViT-L/16",
"family": "Supervised",
"G.PR": 0.11,
"G.Iso": 0.9,
"L.Iso": 0.8,
"JER": 14.9,
"disc": 0.646,
"binding": 0.182
},
{
"model": "BEiTv2",
"family": "Masked",
"G.PR": 0.11,
"G.Iso": 0.9,
"L.Iso": 0.79,
"JER": 19.8,
"disc": 0.714,
"binding": 0.178
},
{
"model": "ConvNeXt Large",
"family": "Supervised",
"G.PR": 0.12,
"G.Iso": 0.89,
"L.Iso": 0.79,
"JER": 14.1,
"disc": 0.702,
"binding": 0.16
},
{
"model": "ViT-B/16",
"family": "Supervised",
"G.PR": 0.19,
"G.Iso": 0.94,
"L.Iso": 0.79,
"JER": 17.1,
"disc": 0.648,
"binding": 0.156
},
{
"model": "ConvNeXt Base",
"family": "Supervised",
"G.PR": 0.0,
"G.Iso": 0.22,
"L.Iso": 0.79,
"JER": 18.7,
"disc": 0.586,
"binding": 0.146
},
{
"model": "CLIP ViT-B/16",
"family": "Vision-Lang",
"G.PR": 0.14,
"G.Iso": 0.91,
"L.Iso": 0.85,
"JER": 20.8,
"disc": 0.5,
"binding": 0.138
},
{
"model": "CLIP ViT-L/14",
"family": "Vision-Lang",
"G.PR": 0.13,
"G.Iso": 0.92,
"L.Iso": 0.86,
"JER": 18.7,
"disc": 0.5,
"binding": 0.13
},
{
"model": "SigLIP ViT-B/16",
"family": "Vision-Lang",
"G.PR": 0.09,
"G.Iso": 0.91,
"L.Iso": 0.83,
"JER": 23.1,
"disc": 0.5,
"binding": 0.128
},
{
"model": "SigLIP SoViT-400M",
"family": "Vision-Lang",
"G.PR": 0.07,
"G.Iso": 0.9,
"L.Iso": 0.82,
"JER": 19.0,
"disc": 0.5,
"binding": 0.128
},
{
"model": "BEiT",
"family": "Masked",
"G.PR": 0.07,
"G.Iso": 0.9,
"L.Iso": 0.72,
"JER": 20.9,
"disc": 0.5,
"binding": 0.126
},
{
"model": "CLIP ViT-B/32",
"family": "Vision-Lang",
"G.PR": 0.14,
"G.Iso": 0.91,
"L.Iso": 0.85,
"JER": 21.7,
"disc": 0.5,
"binding": 0.126
},
{
"model": "EVA-CLIP ViT-E/14",
"family": "Vision-Lang",
"G.PR": 0.08,
"G.Iso": 0.92,
"L.Iso": 0.85,
"JER": 19.4,
"disc": 0.5,
"binding": 0.126
},
{
"model": "EVA-CLIP ViT-B/16",
"family": "Vision-Lang",
"G.PR": 0.14,
"G.Iso": 0.92,
"L.Iso": 0.84,
"JER": 19.0,
"disc": 0.5,
"binding": 0.124
},
{
"model": "EVA-CLIP ViT-L/14",
"family": "Vision-Lang",
"G.PR": 0.1,
"G.Iso": 0.91,
"L.Iso": 0.83,
"JER": 18.1,
"disc": 0.5,
"binding": 0.09
}
]
}

Xet Storage Details

Size:
9.02 kB
·
Xet hash:
9c962331003c696bafbfe4efd3f3459e938739f330aefccd1994d9ed09bb925e

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.