Visual Product Retrieval on Visual Shopping End-to-End Human Evaluation
23.9Extremely Similar Rate@1ViT-B/16 384x
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ViT-B/16 384xPretraining=Annotations-1.3B, Resolution=384x2021.08 | 23.9 | 42.3 | |
| ViT-B/16 224xPretraining=Annotations-1.3B, Resolution=224x2021.08 | 23.6 | 40.6 | |
| ResNeXt-101Pretraining=Annotations-1.3B2021.08 | 19.1 | 38.8 | |
| ResNeXt-101Pretraining=IG-940M2021.08 | 17.9 | 36.8 | |
| ProductionPretraining=IG-940M2021.08 | 17.3 | 39 | |
| ResNeXt-101Pretraining=ImageNet-1k2021.08 | 14.9 | 32.1 |