README.md

---
pipeline_tag: image-classification
license: apache-2.0
base_model: timm/tf_efficientnet_b4.in1k
library_name: zeromodels
tags:
- keras
- zeromodels
- image-classification
- efficientnet
- backbone
- arxiv:1905.11946
- pytorch
- jax
- tf
---

## ***See [our collection](https://huggingface.co/collections/zeromodels/efficientnet-6a8eae835b72b040c1fb77b3) for all versions of EfficientNet.***

# Run EfficientNet 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-EfficientNet-blue)](https://imvision12.github.io/ZeroModels/classification_backbones/) [![Collection](https://img.shields.io/badge/HF-EfficientNet%20collection-yellow)](https://huggingface.co/collections/zeromodels/efficientnet-6a8eae835b72b040c1fb77b3)

# zeromodels/tf_efficientnet_b4_in1k

Paper: [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks (arXiv:1905.11946)](https://arxiv.org/abs/1905.11946) · [HF Papers](https://huggingface.co/papers/1905.11946)

EfficientNet compound-scales depth/width/resolution for strong accuracy/efficiency. Classifier or multi-scale MBConv backbone.

For more details on the model, please go to the upstream [model card](https://huggingface.co/timm/tf_efficientnet_b4.in1k).

Pure-**Keras 3** conversion of [`timm/tf_efficientnet_b4.in1k`](https://huggingface.co/timm/tf_efficientnet_b4.in1k) for [zeromodels](https://github.com/IMvision12/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**.

This is an **image-classification / backbone** checkpoint (`EfficientNetImageClassify` / `EfficientNetModel`).

## ✨ Quick start

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

from PIL import Image
import numpy as np
from zeromodels.models.efficientnet import EfficientNetImageClassify, EfficientNetModel

model = EfficientNetImageClassify.from_weights("zeromodels/tf_efficientnet_b4_in1k")
backbone = EfficientNetModel.from_weights(
    "zeromodels/tf_efficientnet_b4_in1k", 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 EfficientNet variant the same way with `from_weights("zeromodels/<variant>")`:

| Variant | Hub |
|---|---|
| `tf_efficientnet_b0_aa_in1k` | [`zeromodels/tf_efficientnet_b0_aa_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b0_aa_in1k) |
| `tf_efficientnet_b0_ap_in1k` | [`zeromodels/tf_efficientnet_b0_ap_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b0_ap_in1k) |
| `tf_efficientnet_b0_in1k` | [`zeromodels/tf_efficientnet_b0_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b0_in1k) |
| `tf_efficientnet_b0_ns_jft_in1k` | [`zeromodels/tf_efficientnet_b0_ns_jft_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b0_ns_jft_in1k) |
| `tf_efficientnet_b1_aa_in1k` | [`zeromodels/tf_efficientnet_b1_aa_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b1_aa_in1k) |
| `tf_efficientnet_b1_ap_in1k` | [`zeromodels/tf_efficientnet_b1_ap_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b1_ap_in1k) |
| `tf_efficientnet_b1_in1k` | [`zeromodels/tf_efficientnet_b1_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b1_in1k) |
| `tf_efficientnet_b1_ns_jft_in1k` | [`zeromodels/tf_efficientnet_b1_ns_jft_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b1_ns_jft_in1k) |
| `tf_efficientnet_b2_aa_in1k` | [`zeromodels/tf_efficientnet_b2_aa_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b2_aa_in1k) |
| `tf_efficientnet_b2_ap_in1k` | [`zeromodels/tf_efficientnet_b2_ap_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b2_ap_in1k) |
| `tf_efficientnet_b2_in1k` | [`zeromodels/tf_efficientnet_b2_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b2_in1k) |
| `tf_efficientnet_b2_ns_jft_in1k` | [`zeromodels/tf_efficientnet_b2_ns_jft_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b2_ns_jft_in1k) |
| `tf_efficientnet_b3_aa_in1k` | [`zeromodels/tf_efficientnet_b3_aa_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b3_aa_in1k) |
| `tf_efficientnet_b3_ap_in1k` | [`zeromodels/tf_efficientnet_b3_ap_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b3_ap_in1k) |
| `tf_efficientnet_b3_in1k` | [`zeromodels/tf_efficientnet_b3_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b3_in1k) |
| `tf_efficientnet_b3_ns_jft_in1k` | [`zeromodels/tf_efficientnet_b3_ns_jft_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b3_ns_jft_in1k) |
| `tf_efficientnet_b4_aa_in1k` | [`zeromodels/tf_efficientnet_b4_aa_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b4_aa_in1k) |
| `tf_efficientnet_b4_ap_in1k` | [`zeromodels/tf_efficientnet_b4_ap_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b4_ap_in1k) |
| `tf_efficientnet_b4_in1k` | [`zeromodels/tf_efficientnet_b4_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b4_in1k) |
| `tf_efficientnet_b4_ns_jft_in1k` | [`zeromodels/tf_efficientnet_b4_ns_jft_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b4_ns_jft_in1k) |
| `tf_efficientnet_b5_aa_in1k` | [`zeromodels/tf_efficientnet_b5_aa_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b5_aa_in1k) |
| `tf_efficientnet_b5_ap_in1k` | [`zeromodels/tf_efficientnet_b5_ap_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b5_ap_in1k) |
| `tf_efficientnet_b5_in1k` | [`zeromodels/tf_efficientnet_b5_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b5_in1k) |
| `tf_efficientnet_b5_ns_jft_in1k` | [`zeromodels/tf_efficientnet_b5_ns_jft_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b5_ns_jft_in1k) |
| `tf_efficientnet_b6_aa_in1k` | [`zeromodels/tf_efficientnet_b6_aa_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b6_aa_in1k) |
| `tf_efficientnet_b6_ap_in1k` | [`zeromodels/tf_efficientnet_b6_ap_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b6_ap_in1k) |
| `tf_efficientnet_b6_ns_jft_in1k` | [`zeromodels/tf_efficientnet_b6_ns_jft_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b6_ns_jft_in1k) |
| `tf_efficientnet_b7_aa_in1k` | [`zeromodels/tf_efficientnet_b7_aa_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b7_aa_in1k) |
| `tf_efficientnet_b7_ap_in1k` | [`zeromodels/tf_efficientnet_b7_ap_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b7_ap_in1k) |
| `tf_efficientnet_b7_ns_jft_in1k` | [`zeromodels/tf_efficientnet_b7_ns_jft_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b7_ns_jft_in1k) |
| `tf_efficientnet_b8_ap_in1k` | [`zeromodels/tf_efficientnet_b8_ap_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_b8_ap_in1k) |
| `tf_efficientnet_l2_ns_jft_in1k` | [`zeromodels/tf_efficientnet_l2_ns_jft_in1k`](https://huggingface.co/zeromodels/tf_efficientnet_l2_ns_jft_in1k) |
| `tf_efficientnet_l2_ns_jft_in1k_475` | [`zeromodels/tf_efficientnet_l2_ns_jft_in1k_475`](https://huggingface.co/zeromodels/tf_efficientnet_l2_ns_jft_in1k_475) |

## Tips

- Set `KERAS_BACKEND` **before** importing Keras / zeromodels.
- `EfficientNetImageClassify` returns class logits; `EfficientNetModel` 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: `EfficientNetImageClassify.from_weights("hf:timm/tf_efficientnet_b4.in1k")`.

## Special Thanks

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

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