List your company
ResearchFormerly WizWand

Image-Text Retrieval on COCO (val)

60.1R@1DINOv2-ARL
12.509624.864837.2249.5752May 26, 2023Aug 15, 2023Nov 5, 2023Jan 26, 2024Apr 17, 2024Jul 8, 2024Sep 28, 2024
Updated 8mo ago

Evaluation results

MethodLinks
DINOv2-ARLEvaluation Protocol=zero-shot2024.0960.1——
LAION-CLIPBackbone=ViT-L, Evaluation Protocol=zero-shot2024.0959.7——
DINOv2-MpNetEvaluation Protocol=zero-shot2024.0958——
CLIPvisual encoder=ViT-L/14-336, visual FLOPS=191.0G2023.1257.9——
SF-LCLIPvisual encoder=SF-LCLIP, tokens=144, visual FLOPS=33.9G2023.1257——
TinyCLIPvisual encoder=ViT-63M/32-224, visual FLOPS=2.0G2023.1256.9——
CLIPvisual encoder=ViT-L/14-224, visual FLOPS=81.1G2023.1256.3——
OpenAI-CLIPBackbone=ViT-L, Evaluation Protocol=zero-shot2024.0956.3——
SF-LCLIPvisual encoder=SF-LCLIP, tokens=128, visual FLOPS=30.0G2023.1254.7——
3TPretraining=IN-21k, Model Scale=L, Training Dataset=Unfiltered WebLI, Evaluation Protocol=Zero-shot2023.0554.4——
3TPretraining=JFT, Model Scale=L, Training Dataset=Unfiltered WebLI, Evaluation Protocol=Zero-shot2023.0554.4——
LiTPretraining=JFT, Model Scale=L, Training Dataset=Unfiltered WebLI, Evaluation Protocol=Zero-shot2023.0552.7——
CLIPvisual encoder=ViT-B/16-224, visual FLOPS=17.5G2023.1252.4——
SF-LCLIPvisual encoder=SF-LCLIP, tokens=64, visual FLOPS=14.8G2023.1252.3——
Baseline (CLIP/ALIGN)Pretraining=IN-21k, Model Scale=L, Training Dataset=Unfiltered WebLI, Evaluation Protocol=Zero-shot2023.0551——
CLIPvisual encoder=ViT-B/32-224, visual FLOPS=4.4G2023.1250.1——
LiTBackbone=L16L, Evaluation Protocol=zero-shot2024.0948.5——
SF-BCLIPvisual encoder=SF-BCLIP, tokens=49, visual FLOPS=3.8G2023.1247.7——
LiTPretraining=IN-21k, Model Scale=L, Training Dataset=Unfiltered WebLI, Evaluation Protocol=Zero-shot2023.0546.1——
DINOv2-ARLEvaluation Protocol=zero-shot2024.0945.1——
LAION-CLIPBackbone=ViT-L, Evaluation Protocol=zero-shot2024.0943——
DINOv2-MpNetEvaluation Protocol=zero-shot2024.0942.6——
TinyCLIPvisual encoder=ViT-63M/32-224, visual FLOPS=2.0G2023.1238.5——
SF-LCLIPvisual encoder=SF-LCLIP, tokens=144, visual FLOPS=33.9G2023.1238.3——
3TPretraining=JFT, Model Scale=L, Training Dataset=Unfiltered WebLI, Evaluation Protocol=Zero-shot2023.0537.9——
3TPretraining=IN-21k, Model Scale=L, Training Dataset=Unfiltered WebLI, Evaluation Protocol=Zero-shot2023.0537.7——
CLIPvisual encoder=ViT-L/14-336, visual FLOPS=191.0G2023.1237.1——
LiTPretraining=JFT, Model Scale=L, Training Dataset=Unfiltered WebLI, Evaluation Protocol=Zero-shot2023.0536.7——
CLIPvisual encoder=ViT-L/14-224, visual FLOPS=81.1G2023.1236.5——
OpenAI-CLIPBackbone=ViT-L, Evaluation Protocol=zero-shot2024.0936.5——
SF-LCLIPvisual encoder=SF-LCLIP, tokens=128, visual FLOPS=30.0G2023.1236.4——
SF-LCLIPvisual encoder=SF-LCLIP, tokens=64, visual FLOPS=14.8G2023.1234.8——
Baseline (CLIP/ALIGN)Pretraining=IN-21k, Model Scale=L, Training Dataset=Unfiltered WebLI, Evaluation Protocol=Zero-shot2023.0534.2——
CLIPvisual encoder=ViT-B/16-224, visual FLOPS=17.5G2023.1233——
CLIPzero-shot=true2023.1032.5844.5757.62
Ex-MCRzero-shot=true2023.1032.5844.5757.62
SF-BCLIPvisual encoder=SF-BCLIP, tokens=49, visual FLOPS=3.8G2023.1231.5——
LiTBackbone=L16L, Evaluation Protocol=zero-shot2024.0931.2——
CLIPvisual encoder=ViT-B/32-224, visual FLOPS=4.4G2023.1230.4——
LiTPretraining=IN-21k, Model Scale=L, Training Dataset=Unfiltered WebLI, Evaluation Protocol=Zero-shot2023.0527.8——
ULIPzero-shot=true2023.1022.9234.4246.33
ULIP v2zero-shot=true2023.1022.9234.4246.33
C-MCRzero-shot=true2023.1014.3424.2333.19
AudioCLIPZero-shot=true2023.10—20.1427.42
C-MCRZero-shot=true2023.10—24.5633.83
CLIPZero-shot=true2023.10—44.5757.62
Ex-MCRZero-shot=true2023.10—44.5757.62
WAV2CLIPZero-shot=true2023.10—44.5757.62