examples/README.md

# Examples

## Zero-Shot Classification

```sh
pip install torch  # whatever version you want
pip install open_clip_torch numpy tqdm torchvision
```

Suppose you want to evaluate BioCLIP on zero-shot classification on two tasks, `<DATASET-NAME>` and `<DATASET2-NAME>`.
You can use `examples/zero_shot.py` to get top1 and top5 accuracy assuming your tasks are arranged as `torchvision`'s [`ImageFolder`](https://pytorch.org/vision/stable/generated/torchvision.datasets.ImageFolder.html) wants.

```sh
python examples/zero_shot.py \
  --datasets <DATASET-NAME>=<DATASET-FOLDER> <DATASET2-NAME>=<DATASET2-FOLDER>
```

This will write to `logs/bioclip-zero-shot/results.json` with your results.