List your company

Zero-Shot Classification Suite on CLIP Standard (test val)

45ImageNet AccuracyRILS
33.97636.83839.742.562Jan 17, 2023Apr 7, 2023Jun 27, 2023Sep 15, 2023Dec 5, 2023Feb 23, 2024May 14, 2024

Evaluation results

MethodLinks
RILSEvaluation Protocol=Zero-shot, Backbone=ViT-B/16, Training Data=L-20M, Training Epochs=252023.014558.986.255.123.451.819.55.932.859.280.733.522.640.193.228.840.219.17.816.85042.317—————
MAE+CLIPEvaluation Protocol=Zero-shot, Backbone=ViT-B/16, Training Data=L-20M, Training Epochs=252023.0142.357.878.252.423.951.618.14.631.555.878.43227.632.789.82739.422.97.214.749.339.90—————
SLIPEvaluation Protocol=Zero-shot, Backbone=ViT-B/16, Training Data=L-20M, Training Epochs=252023.0141.656.773.443.222.651.617.74.932.452.579.133.329.433.589.517.836.217.86.823.449.738.72—————
CLIPEvaluation Protocol=Zero-shot, Backbone=ViT-B/16, Training Data=L-20M, Training Epochs=252023.0140.355.77646.924.450.717.84.831.553.778.431.826.837.68922.736.924.16.82049.139.32—————
Ours-RGB0.3Zero-shot=true, Normalized Training Time (N.T.)=0.64x2024.0536.6—63.126.3—————57.372.325.49.4————————36.8—8.839.945.924.931.8
Ours-EmbeddingZero-shot=true, Normalized Training Time (N.T.)=0.64x2024.0536.3—70.732—————55.473.728.411————————38.2—8.139.647.925.430.7
CLIPZero-shot=true, Normalized Training Time (N.T.)=1.00x2024.0536.1—57.725—————53.973.52611.7————————36.1—838.447.624.930.7
Ours-RGB0.5Zero-shot=true, Normalized Training Time (N.T.)=0.54x2024.0536—63.627.2—————5572.226.111.2————————36.5—7.739.845.323.830.5
Ours-KmeansZero-shot=true, Normalized Training Time (N.T.)=0.70x2024.0535.5—61.425.3—————52.169.925.69.7————————35—7.238.34322.230.1
FLIP AttnZero-shot=true, Normalized Training Time (N.T.)=1.06x2024.0535.2—61.627.1—————51.273.525.79.4————————35.8—8.139.445.123.730.1
FLIPZero-shot=true, Normalized Training Time (N.T.)=0.53x2024.0534.4—52.824.5—————4670.425.310.4————————33.5—7.139.541.420.129.5