|
|
--- |
|
|
license: mit |
|
|
--- |
|
|
|
|
|
# Classifiers Enhanced by Pre-training |
|
|
|
|
|
This project utilizes a visual encoder from the pre-trained CLIP (ViT-B/32) to build image classifiers. To use the trained models, follow the steps below to set up and run the classifiers. |
|
|
|
|
|
## Prerequisites |
|
|
|
|
|
Before you start, make sure you have Python and the necessary libraries installed. |
|
|
|
|
|
## Download the Trained Models and CIFAR-100 Dataset |
|
|
|
|
|
You need to download the following trained model weights and CIFAR-100 dataset for running the project: |
|
|
- `fine-tune-best.pth`: Best model weights after fine-tuning. |
|
|
- `linear-probe-best.pth`: Best model weights after the linear probe training. |
|
|
- `train-from-scratch-best.pth`: Best model weights trained from scratch. |
|
|
|
|
|
Please download these files and place them under the `results/` directory within the project folder. |
|
|
|
|
|
- `cifar-100-python.tar.gz`: CIFAR-100 dataset. |
|
|
|
|
|
Please download this file and place it under the `data/` directory within the project folder. |
|
|
|
|
|
|
|
|
## Installation and Usage |
|
|
|
|
|
See https://github.com/Gengsheng-Li/Classifiers-enhanced-by-pre-training for more details. |
|
|
|
