README.md

---
pipeline_tag: image-classification
license: apache-2.0
base_model: timm/tf_efficientnetv2_l.in21k
library_name: zeromodels
tags:
- keras
- zeromodels
- image-classification
- efficientnetv2
- backbone
- arxiv:2104.00298
- pytorch
- jax
- tf
---

## ***See [our collection](https://huggingface.co/collections/zeromodels/efficientnetv2-6a8eae77918224be1104752f) for all versions of EfficientNetV2.***

# Run EfficientNetV2 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-EfficientNetV2-blue)](https://imvision12.github.io/ZeroModels/classification_backbones/) [![Collection](https://img.shields.io/badge/HF-EfficientNetV2%20collection-yellow)](https://huggingface.co/collections/zeromodels/efficientnetv2-6a8eae77918224be1104752f)

# zeromodels/tf_efficientnetv2_l_in21k

Paper: [EfficientNetV2: Smaller Models and Faster Training (arXiv:2104.00298)](https://arxiv.org/abs/2104.00298) · [HF Papers](https://huggingface.co/papers/2104.00298)

EfficientNetV2 trains faster with Fused-MBConv and progressive learning. Same ImageClassify / Model API as EfficientNet.

For more details on the model, please go to the upstream [model card](https://huggingface.co/timm/tf_efficientnetv2_l.in21k).

Pure-**Keras 3** conversion of [`timm/tf_efficientnetv2_l.in21k`](https://huggingface.co/timm/tf_efficientnetv2_l.in21k) for [zeromodels](https://github.com/IMvision12/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**.

This is an **image-classification / backbone** checkpoint (`EfficientNetV2ImageClassify` / `EfficientNetV2Model`).

## ✨ Quick start

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

from PIL import Image
import numpy as np
from zeromodels.models.efficientnetv2 import EfficientNetV2ImageClassify, EfficientNetV2Model

model = EfficientNetV2ImageClassify.from_weights("zeromodels/tf_efficientnetv2_l_in21k")
backbone = EfficientNetV2Model.from_weights(
    "zeromodels/tf_efficientnetv2_l_in21k", as_backbone=True
)

image = Image.open("your_image.jpg").convert("RGB")
image = image.resize((224, 224))
x = np.asarray(image, dtype="float32")[None]  # (1, H, W, 3)
print(model(x).shape)  # (1, num_classes)
feats = backbone(x)
print(len(feats), [tuple(f.shape) for f in feats])
```

Load any EfficientNetV2 variant the same way with `from_weights("zeromodels/<variant>")`:

| Variant | Hub |
|---|---|
| `tf_efficientnetv2_b0_in1k` | [`zeromodels/tf_efficientnetv2_b0_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_b0_in1k) |
| `tf_efficientnetv2_b1_in1k` | [`zeromodels/tf_efficientnetv2_b1_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_b1_in1k) |
| `tf_efficientnetv2_b2_in1k` | [`zeromodels/tf_efficientnetv2_b2_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_b2_in1k) |
| `tf_efficientnetv2_b3_in1k` | [`zeromodels/tf_efficientnetv2_b3_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_b3_in1k) |
| `tf_efficientnetv2_b3_in21k_ft_in1k` | [`zeromodels/tf_efficientnetv2_b3_in21k_ft_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_b3_in21k_ft_in1k) |
| `tf_efficientnetv2_l_in1k` | [`zeromodels/tf_efficientnetv2_l_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_l_in1k) |
| `tf_efficientnetv2_l_in21k` | [`zeromodels/tf_efficientnetv2_l_in21k`](https://huggingface.co/zeromodels/tf_efficientnetv2_l_in21k) |
| `tf_efficientnetv2_l_in21k_ft_in1k` | [`zeromodels/tf_efficientnetv2_l_in21k_ft_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_l_in21k_ft_in1k) |
| `tf_efficientnetv2_m_in1k` | [`zeromodels/tf_efficientnetv2_m_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_m_in1k) |
| `tf_efficientnetv2_m_in21k` | [`zeromodels/tf_efficientnetv2_m_in21k`](https://huggingface.co/zeromodels/tf_efficientnetv2_m_in21k) |
| `tf_efficientnetv2_m_in21k_ft_in1k` | [`zeromodels/tf_efficientnetv2_m_in21k_ft_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_m_in21k_ft_in1k) |
| `tf_efficientnetv2_s_in1k` | [`zeromodels/tf_efficientnetv2_s_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_s_in1k) |
| `tf_efficientnetv2_s_in21k` | [`zeromodels/tf_efficientnetv2_s_in21k`](https://huggingface.co/zeromodels/tf_efficientnetv2_s_in21k) |
| `tf_efficientnetv2_s_in21k_ft_in1k` | [`zeromodels/tf_efficientnetv2_s_in21k_ft_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_s_in21k_ft_in1k) |
| `tf_efficientnetv2_xl_in21k` | [`zeromodels/tf_efficientnetv2_xl_in21k`](https://huggingface.co/zeromodels/tf_efficientnetv2_xl_in21k) |
| `tf_efficientnetv2_xl_in21k_ft_in1k` | [`zeromodels/tf_efficientnetv2_xl_in21k_ft_in1k`](https://huggingface.co/zeromodels/tf_efficientnetv2_xl_in21k_ft_in1k) |

## Tips

- Set `KERAS_BACKEND` **before** importing Keras / zeromodels.
- `EfficientNetV2ImageClassify` returns class logits; `EfficientNetV2Model` 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: `EfficientNetV2ImageClassify.from_weights("hf:timm/tf_efficientnetv2_l.in21k")`.

## Special Thanks

A huge thank you to the EfficientNetV2 authors and the timm / Hub communities for creating and releasing these models.

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