---
pipeline_tag: image-classification
license: apache-2.0
base_model: timm/densenet201.tv_in1k
library_name: zeromodels
tags:
- keras
- zeromodels
- image-classification
- densenet
- backbone
- arxiv:1608.06993
- pytorch
- jax
- tf
---

## ***See [our collection](https://huggingface.co/collections/zeromodels/densenet-6a8eaefa55ddd8df0fc59650) for all versions of DenseNet.***

# Run DenseNet 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-DenseNet-blue)](https://imvision12.github.io/ZeroModels/classification_backbones/) [![Collection](https://img.shields.io/badge/HF-DenseNet%20collection-yellow)](https://huggingface.co/collections/zeromodels/densenet-6a8eaefa55ddd8df0fc59650)

# zeromodels/densenet201_tv_in1k

Paper: [Densely Connected Convolutional Networks (arXiv:1608.06993)](https://arxiv.org/abs/1608.06993) · [HF Papers](https://huggingface.co/papers/1608.06993)

DenseNet connects each layer to every other layer in a feed-forward fashion. Classifier or dense-block backbone.

For more details on the model, please go to the upstream [model card](https://huggingface.co/timm/densenet201.tv_in1k).

Pure-**Keras 3** conversion of [`timm/densenet201.tv_in1k`](https://huggingface.co/timm/densenet201.tv_in1k) for [zeromodels](https://github.com/IMvision12/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**.

This is an **image-classification / backbone** checkpoint (`DenseNetImageClassify` / `DenseNetModel`).

## ✨ Quick start

```python
import os

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

from PIL import Image
from zeromodels.models.densenet import DenseNetImageClassify, DenseNetModel, DenseNetImageProcessor

model = DenseNetImageClassify.from_weights("zeromodels/densenet201_tv_in1k")
processor = DenseNetImageProcessor.from_weights("zeromodels/densenet201_tv_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 = DenseNetModel.from_weights("zeromodels/densenet201_tv_in1k", as_backbone=True)
features = backbone(pixels, training=False)
```

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

| Variant | Hub |
|---|---|
| `densenet121_tv_in1k` | [`zeromodels/densenet121_tv_in1k`](https://huggingface.co/zeromodels/densenet121_tv_in1k) |
| `densenet161_tv_in1k` | [`zeromodels/densenet161_tv_in1k`](https://huggingface.co/zeromodels/densenet161_tv_in1k) |
| `densenet169_tv_in1k` | [`zeromodels/densenet169_tv_in1k`](https://huggingface.co/zeromodels/densenet169_tv_in1k) |
| `densenet201_tv_in1k` | [`zeromodels/densenet201_tv_in1k`](https://huggingface.co/zeromodels/densenet201_tv_in1k) |

## Tips

- Set `KERAS_BACKEND` **before** importing Keras / zeromodels.
- `DenseNetImageClassify` returns class logits; `DenseNetModel` 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: `DenseNetImageClassify.from_weights("hf:timm/densenet201.tv_in1k")`.

## Special Thanks

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

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