Benchmarks

Task NameDataset NameSOTA ResultTrendResults

Scene Recognition

PLACES-LT ResNet-152 (test)

45.5Hit@1

Image Inpainting

Places 256 x 256 (Standard)

1.02FID (Small Mask)

Image Classification

Places205 (test)

60.4Accuracy

Single Image Generation

Places50 (test)

0.06SIFID

Inpainting (Center 50% x 50%)

Places2

7.13FID

Outpainting (Right 50%)

Places2

6.78FID

Global Colourisation

Place365 Outdoor

129.11FID

Long-tail Recognition

Places-LT

45.5Conditional Hit@1

Image Classification

Places365

35Accuracy (Seen)

Long-tailed Image Classification

Places-LT

53.7Overall Accuracy

Single Image Generation

Places 50

81Preference Score

Audio-to-image retrieval

Places 400k

53.4R@1

Image Classification

Places-365 256 x 256 (test)

49.83Top-1 Accuracy

Random Image Generation

Places 50 365 (Evaluation)

0.044SIFID

Image Completion

Places2 512x512 (val)

1,330P-IDS

Image Inpainting

Places (val)

0.057L1 Distance

Image Inpainting

Places365 Inpainting (held-out set of 500 images)

2.53FID

Image Extension

Places365 75% extension (held-out set of 500 images)

8.79FID

Image Extension

Places365 held-out set of 500 images (50% extension)

3.46FID

Image Extension

Places365 25% extension (held-out set of 500 images)

0.79FID

Long-tailed Visual Recognition

Places-LT

45.3Acc (Many)

Symbolic Distillation

Places

0.077Mean R^2

Out-of-Distribution Detection

Places OOD

99.82AUPR

Uncertainty Calibration

Places365 OOD (test)

48.65RMS Calibration Error

Scene Classification

Places-LT 1.0 (test)

38.66Top-1 Accuracy