BiT is a cutting-edge set of pre-trained image transfer learning models in the field of computer vision: Even with only a small number of samples per class, it can achieve excellent performance on new data sets after migration
| Model | Download | Top-1 accuracy (%) | Top-5 accuracy (%) |
|---|---|---|---|
| BiT-M-R50x1 | https://storage.googleapis.com/bit_models/BiT-M-R50x1.npz | 18.36 | 50.64 |
- Big Transfer model was described in the paper titled [Big Transfer (BiT): General Visual Representation Learning] (https://arxiv.org/abs/1912.11370)
- The pre-train model is from the Project: (https://github.com/google-research/big_transfer).
Apache 2.0: