# HRNet

**Category:** CNN · **Code:**
[`timm/models/hrnet.py`](https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/hrnet.py) ·
**Architectures:** 11 · **Pretrained weights:** 14 (14 with ImageNet-1k results)

High-Resolution Networks keep a high resolution branch through the whole network and add lower resolution branches in
parallel, repeatedly exchanging information between them.

## Papers

* [Deep High-Resolution Representation Learning for Visual Recognition](https://arxiv.org/abs/1908.07919)
* [Beyond Self-Supervision: A Simple Yet Effective Network Distillation Alternative to Improve
  Backbones](https://arxiv.org/abs/2103.05959)

## Usage

See the [Quickstart](../quickstart) for loading pretrained weights, image preprocessing and fine-tuning, and [Feature
Extraction](../feature_extraction) for using the models as backbones.

```py
>>> import timm
>>> model = timm.create_model('hrnet_w18.ms_aug_in1k', pretrained=True)
>>> data_config = timm.data.resolve_model_data_config(model)
```

## Model API support

| Feature | `HighResolutionNet` |
|---|---|
| [`features_only=True` feature maps](../model_api#intermediate-features) | ✓ (strides 2, 4, 8, 16, 32) |
| [`forward_intermediates()`](../model_api#intermediate-features) | ✗ |
| [`prune_intermediate_layers()`](../model_api#intermediate-features) | ✗ |
| [Any input size, no changes needed](../model_api#input-size) | ✓ |
| [`set_input_size()`](../model_api#input-size) | – |
| [`dynamic_img_size=True`](../model_api#input-size) | – |
| [`set_grad_checkpointing()`](../model_api#training-hooks) | ✗ |
| [Layer-wise LR decay (`group_matcher()`)](../model_api#training-hooks) | ✓ |
| [`reset_classifier()`](../model_api#classifier) | ✓ |

✓ supported, ✓ (square) works for square sizes only, ✗ not supported, – not needed (the model already accepts any input
size). Checked by running each feature on a randomly initialized model (`HighResolutionNet` with `hrnet_w18_small` at
224), other architectures of the same class are expected to behave the same. See the [Model API](../model_api) page for
details.

## Architectures

| Architecture | Params (M) | GMACs | Input size | Pretrained weights |
|---|---:|---:|---:|---:|
| `hrnet_w18` | 21.30 | 4.32 | 224 | 2 |
| `hrnet_w18_small` | 13.19 | 1.61 | 224 | 2 |
| `hrnet_w18_small_v2` | 15.60 | 2.62 | 224 | 2 |
| `hrnet_w18_ssld` | 21.30 | 4.32 | 224 | 1 |
| `hrnet_w30` | 37.71 | 8.15 | 224 | 1 |
| `hrnet_w32` | 41.23 | 8.97 | 224 | 1 |
| `hrnet_w40` | 57.56 | 12.75 | 224 | 1 |
| `hrnet_w44` | 67.06 | 14.94 | 224 | 1 |
| `hrnet_w48` | 77.47 | 17.34 | 224 | 1 |
| `hrnet_w48_ssld` | 77.47 | 17.34 | 224 | 1 |
| `hrnet_w64` | 128.06 | 28.97 | 224 | 1 |

GMACs and parameter counts are for the default input size and classifier of each architecture, from the [benchmark
results](https://github.com/huggingface/pytorch-image-models/tree/main/results).

## Pretrained weights

Top-1 is ImageNet-1k validation accuracy from
[`results-imagenet.csv`](https://github.com/huggingface/pytorch-image-models/blob/main/results/results-imagenet.csv) at
the listed eval size, - if the weight has no ImageNet-1k result (e.g. pretraining or image-text weights). Classes is the
size of the pretrained classifier, none for weights without one.

Weights are grouped by source: the paper or origin given in the pretrained config, else a Hub organization other than
`timm`, else the first part of the pretrained tag (`architecture.tag`). Sources with a single weight are listed under
other.

### `ms` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `hrnet_w18.ms_aug_in1k` | 224 | 78.110 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w18.ms_aug_in1k) |
| `hrnet_w18.ms_in1k` | 224 | 76.770 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w18.ms_in1k) |
| `hrnet_w18_small.ms_in1k` | 224 | 72.340 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w18_small.ms_in1k) |
| `hrnet_w18_small_v2.ms_in1k` | 224 | 75.094 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w18_small_v2.ms_in1k) |
| `hrnet_w30.ms_in1k` | 224 | 78.202 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w30.ms_in1k) |
| `hrnet_w32.ms_in1k` | 224 | 78.452 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w32.ms_in1k) |
| `hrnet_w40.ms_in1k` | 224 | 78.928 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w40.ms_in1k) |
| `hrnet_w44.ms_in1k` | 224 | 78.890 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w44.ms_in1k) |
| `hrnet_w48.ms_in1k` | 224 | 79.328 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w48.ms_in1k) |
| `hrnet_w64.ms_in1k` | 224 | 79.458 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w64.ms_in1k) |

### `gluon` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `hrnet_w18_small.gluon_in1k` | 224 | 73.934 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/hrnet_w18_small.gluon_in1k) |
| `hrnet_w18_small_v2.gluon_in1k` | 224 | 78.202 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/hrnet_w18_small_v2.gluon_in1k) |

### `paddle` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `hrnet_w18_ssld.paddle_in1k` | 224 (test 288) | 82.052 | 288 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w18_ssld.paddle_in1k) |
| `hrnet_w48_ssld.paddle_in1k` | 224 (test 288) | 84.484 | 288 | 1000 | mit | [timm](https://huggingface.co/timm/hrnet_w48_ssld.paddle_in1k) |