---
pipeline_tag: image-classification
license: mit
base_model: timm/swinv2_tiny_window16_256.ms_in1k
library_name: zeromodels
tags:
- keras
- zeromodels
- image-classification
- swinv2
- backbone
- arxiv:2111.09883
- pytorch
- jax
- tf
---

## ***See [our collection](https://huggingface.co/collections/zeromodels/swin-transformer-v2-6a8eae9598bdcc4b6c59f10a) for all versions of Swin Transformer V2.***

# Run Swin Transformer V2 with Keras 3: JAX, PyTorch, or TensorFlow

[![GitHub](https://img.shields.io/badge/GitHub-ZeroModels-black?logo=github)](https://github.com/IMvision12/ZeroModels) [![Docs](https://img.shields.io/badge/Docs-Swin--V2-blue)](https://imvision12.github.io/ZeroModels/classification_backbones/) [![Collection](https://img.shields.io/badge/HF-Swin--V2%20collection-yellow)](https://huggingface.co/collections/zeromodels/swin-transformer-v2-6a8eae9598bdcc4b6c59f10a)

# zeromodels/swinv2_tiny_window16_256_ms_in1k

Paper: [Swin Transformer V2: Scaling Up Capacity and Resolution (arXiv:2111.09883)](https://arxiv.org/abs/2111.09883) · [HF Papers](https://huggingface.co/papers/2111.09883)

Swin V2 scales capacity and resolution with residual post-norm and scaled cosine attention. Classifier + hierarchical backbone.

For more details on the model, please go to the upstream [model card](https://huggingface.co/timm/swinv2_tiny_window16_256.ms_in1k).

Pure-**Keras 3** conversion of [`timm/swinv2_tiny_window16_256.ms_in1k`](https://huggingface.co/timm/swinv2_tiny_window16_256.ms_in1k) for [zeromodels](https://github.com/IMvision12/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**.

This is an **image-classification / backbone** checkpoint (`SwinV2ImageClassify` / `SwinV2Model`).

## ✨ Quick start

```python
import os

os.environ["KERAS_BACKEND"] = "torch"  # or "jax" / "tensorflow"

from PIL import Image
from zeromodels.models.swinv2 import SwinV2ImageClassify, SwinV2Model, SwinV2ImageProcessor

model = SwinV2ImageClassify.from_weights("zeromodels/swinv2_tiny_window16_256_ms_in1k")
processor = SwinV2ImageProcessor.from_weights("zeromodels/swinv2_tiny_window16_256_ms_in1k")

image = Image.open("your_image.jpg").convert("RGB")
pixels = processor(image)  # resize + normalize (normalization lives in the processor)
logits = model(pixels, training=False)
print(logits.shape)  # (1, num_classes)

# Feature extraction: the backbone without the classifier head
backbone = SwinV2Model.from_weights("zeromodels/swinv2_tiny_window16_256_ms_in1k", as_backbone=True)
features = backbone(pixels, training=False)
```

Load any Swin Transformer V2 variant the same way with `from_weights("zeromodels/<variant>")`:

| Variant | Hub |
|---|---|
| `swinv2_base_window12_192_ms_in22k` | [`zeromodels/swinv2_base_window12_192_ms_in22k`](https://huggingface.co/zeromodels/swinv2_base_window12_192_ms_in22k) |
| `swinv2_base_window12to16_192to256_ms_in22k_ft_in1k` | [`zeromodels/swinv2_base_window12to16_192to256_ms_in22k_ft_in1k`](https://huggingface.co/zeromodels/swinv2_base_window12to16_192to256_ms_in22k_ft_in1k) |
| `swinv2_base_window12to24_192to384_ms_in22k_ft_in1k` | [`zeromodels/swinv2_base_window12to24_192to384_ms_in22k_ft_in1k`](https://huggingface.co/zeromodels/swinv2_base_window12to24_192to384_ms_in22k_ft_in1k) |
| `swinv2_base_window16_256_ms_in1k` | [`zeromodels/swinv2_base_window16_256_ms_in1k`](https://huggingface.co/zeromodels/swinv2_base_window16_256_ms_in1k) |
| `swinv2_base_window8_256_ms_in1k` | [`zeromodels/swinv2_base_window8_256_ms_in1k`](https://huggingface.co/zeromodels/swinv2_base_window8_256_ms_in1k) |
| `swinv2_large_window12_192_ms_in22k` | [`zeromodels/swinv2_large_window12_192_ms_in22k`](https://huggingface.co/zeromodels/swinv2_large_window12_192_ms_in22k) |
| `swinv2_large_window12to16_192to256_ms_in22k_ft_in1k` | [`zeromodels/swinv2_large_window12to16_192to256_ms_in22k_ft_in1k`](https://huggingface.co/zeromodels/swinv2_large_window12to16_192to256_ms_in22k_ft_in1k) |
| `swinv2_large_window12to24_192to384_ms_in22k_ft_in1k` | [`zeromodels/swinv2_large_window12to24_192to384_ms_in22k_ft_in1k`](https://huggingface.co/zeromodels/swinv2_large_window12to24_192to384_ms_in22k_ft_in1k) |
| `swinv2_small_window16_256_ms_in1k` | [`zeromodels/swinv2_small_window16_256_ms_in1k`](https://huggingface.co/zeromodels/swinv2_small_window16_256_ms_in1k) |
| `swinv2_small_window8_256_ms_in1k` | [`zeromodels/swinv2_small_window8_256_ms_in1k`](https://huggingface.co/zeromodels/swinv2_small_window8_256_ms_in1k) |
| `swinv2_tiny_window16_256_ms_in1k` | [`zeromodels/swinv2_tiny_window16_256_ms_in1k`](https://huggingface.co/zeromodels/swinv2_tiny_window16_256_ms_in1k) |
| `swinv2_tiny_window8_256_ms_in1k` | [`zeromodels/swinv2_tiny_window8_256_ms_in1k`](https://huggingface.co/zeromodels/swinv2_tiny_window8_256_ms_in1k) |

## Tips

- Set `KERAS_BACKEND` **before** importing Keras / zeromodels.
- `SwinV2ImageClassify` returns class logits; `SwinV2Model` returns features (`as_backbone=True` for multi-scale stages).
- See [docs](https://imvision12.github.io/ZeroModels/classification_backbones/) and [Loading Weights](https://imvision12.github.io/ZeroModels/loading_weights/).
- Upstream / timm checkpoints: `SwinV2ImageClassify.from_weights("hf:timm/swinv2_tiny_window16_256.ms_in1k")`.

## Special Thanks

A huge thank you to the Swin Transformer V2 authors and the timm / Hub communities for creating and releasing these models.

License: see YAML `license` (usually matches the upstream checkpoint).