---
library_name: lucid
license: apache-2.0
tags:
  - image-classification
  - cvt
  - lucid
datasets:
  - imagenet-1k
pipeline_tag: image-classification
model-index:
  - name: cvt-21
    results:
      - task: { type: image-classification }
        dataset: { name: ImageNet-1k, type: imagenet-1k }
        metrics:
          - { type: acc@1, value: 82.5 }
---

# CvT-21

> Wu et al., 2021 — *CvT: Introducing Convolutions to Vision Transformers* (arXiv:2103.15808)

[Lucid](https://github.com/ChanLumerico/lucid) port of `transformers/microsoft/cvt-21`,
converted to Lucid-native safetensors.

## Available weights

| Tag | acc@1 | acc@5 | Params | GFLOPs | Size | Source |
|---|---|---|---|---|---|---|
| `IN1K` *(default)* | 82.5 | — | 31.6M | — | 120.87 MB | transformers |

## Usage

```python
import lucid.models as models
from lucid.models.weights import CvT21Weights

# default tag
model = models.cvt_21_cls(pretrained=True)

# explicit tag (enum or string)
model = models.cvt_21_cls(weights=CvT21Weights.IN1K)
model = models.cvt_21_cls(pretrained="IN1K")

# preprocessing travels with the weights
weights = CvT21Weights.IN1K
preprocess = weights.transforms()
logits = model(preprocess(image)[None]).logits
```

## Conversion

Converted from `transformers/microsoft/cvt-21` via
`python -m tools.convert_weights cvt_21 --tag IN1K`.
Key mapping + numerical parity verified against the source.

## License

`apache-2.0` — inherited from the original weights.

## Citation

```
@inproceedings{wu2021cvt,
  title={CvT: Introducing Convolutions to Vision Transformers},
  author={Wu, Haiping and Xiao, Bin and Codella, Noel and Liu, Mengchen and Dai, Xiyang and Yuan, Lu and Zhang, Lei},
  booktitle={ICCV}, year={2021}
}
```