# RegNet

**Category:** CNN · **Code:**
[`timm/models/regnet.py`](https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/regnet.py) ·
**Architectures:** 36 · **Pretrained weights:** 65 (60 with ImageNet-1k results)

RegNetX and RegNetY networks from the regular network design spaces, RegNetZ from "Fast and Accurate Model Scaling", and
`timm` additions such as the pre-activation RegNetV. Includes weights from pycls, torchvision and VISSL (SEER).

## Papers

* [Designing Network Design Spaces](https://arxiv.org/abs/2003.13678)
* [Fast and Accurate Model Scaling](https://arxiv.org/abs/2103.06877)
* [Self-supervised Pretraining of Visual Features in the Wild](https://arxiv.org/abs/2103.01988)
* [Revisiting Weakly Supervised Pre-Training of Visual Perception Models](https://arxiv.org/abs/2201.08371)

## 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('regnetx_064.pycls_in1k', pretrained=True)
>>> data_config = timm.data.resolve_model_data_config(model)
```

## Model API support

| Feature | `RegNet` |
|---|---|
| [`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 (`RegNet` with `regnetx_002` 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 |
|---|---:|---:|---:|---:|
| `regnetv_040` | 20.64 | 4.00 | 224 | 1 |
| `regnetv_064` | 30.58 | 6.39 | 224 | 1 |
| `regnetx_002` | 2.68 | 0.20 | 224 | 1 |
| `regnetx_004` | 5.16 | 0.40 | 224 | 1 |
| `regnetx_004_tv` | 5.50 | 0.42 | 224 | 1 |
| `regnetx_006` | 6.20 | 0.61 | 224 | 1 |
| `regnetx_008` | 7.26 | 0.81 | 224 | 2 |
| `regnetx_016` | 9.19 | 1.62 | 224 | 2 |
| `regnetx_032` | 15.30 | 3.20 | 224 | 2 |
| `regnetx_040` | 22.12 | 3.99 | 224 | 1 |
| `regnetx_064` | 26.21 | 6.49 | 224 | 1 |
| `regnetx_080` | 39.57 | 8.02 | 224 | 2 |
| `regnetx_120` | 46.11 | 12.13 | 224 | 1 |
| `regnetx_160` | 54.28 | 15.99 | 224 | 2 |
| `regnetx_320` | 107.81 | 31.81 | 224 | 2 |
| `regnety_002` | 3.16 | 0.20 | 224 | 1 |
| `regnety_004` | 4.34 | 0.41 | 224 | 2 |
| `regnety_006` | 6.06 | 0.61 | 224 | 1 |
| `regnety_008` | 6.26 | 0.81 | 224 | 1 |
| `regnety_008_tv` | 6.43 | 0.84 | 224 | 1 |
| `regnety_016` | 11.20 | 1.63 | 224 | 2 |
| `regnety_032` | 19.44 | 3.20 | 224 | 3 |
| `regnety_040` | 20.65 | 4.00 | 224 | 2 |
| `regnety_040_sgn` | 20.65 | 4.03 | 224 | 0 |
| `regnety_064` | 30.58 | 6.39 | 224 | 2 |
| `regnety_080` | 39.18 | 8.00 | 224 | 2 |
| `regnety_080_tv` | 39.38 | 8.51 | 224 | 1 |
| `regnety_120` | 51.82 | 12.14 | 224 | 3 |
| `regnety_160` | 83.59 | 15.96 | 224 | 8 |
| `regnety_320` | 145.05 | 32.34 | 224 | 6 |
| `regnety_640` | 281.38 | 188.47 | 384 | 2 |
| `regnety_1280` | 644.81 | 374.99 | 384 | 4 |
| `regnety_2560` | 1282.60 | 747.83 | 384 | 1 |
| `regnetz_005` | 7.12 | 0.52 | 224 | 0 |
| `regnetz_040` | 27.12 | 4.06 | 256 | 1 |
| `regnetz_040_h` | 28.94 | 4.12 | 256 | 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.

### facebookresearch/pycls

Origin: [github.com/facebookresearch/pycls](https://github.com/facebookresearch/pycls)

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `regnetx_002.pycls_in1k` | 224 | 68.758 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_002.pycls_in1k) |
| `regnetx_004.pycls_in1k` | 224 | 72.482 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_004.pycls_in1k) |
| `regnetx_006.pycls_in1k` | 224 | 73.904 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_006.pycls_in1k) |
| `regnetx_008.pycls_in1k` | 224 | 75.054 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_008.pycls_in1k) |
| `regnetx_016.pycls_in1k` | 224 | 76.936 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_016.pycls_in1k) |
| `regnetx_032.pycls_in1k` | 224 | 78.160 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_032.pycls_in1k) |
| `regnetx_040.pycls_in1k` | 224 | 78.486 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_040.pycls_in1k) |
| `regnetx_064.pycls_in1k` | 224 | 79.062 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_064.pycls_in1k) |
| `regnetx_080.pycls_in1k` | 224 | 79.214 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_080.pycls_in1k) |
| `regnetx_120.pycls_in1k` | 224 | 79.608 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_120.pycls_in1k) |
| `regnetx_160.pycls_in1k` | 224 | 79.844 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_160.pycls_in1k) |
| `regnetx_320.pycls_in1k` | 224 | 80.258 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnetx_320.pycls_in1k) |
| `regnety_002.pycls_in1k` | 224 | 70.288 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_002.pycls_in1k) |
| `regnety_004.pycls_in1k` | 224 | 74.004 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_004.pycls_in1k) |
| `regnety_006.pycls_in1k` | 224 | 75.274 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_006.pycls_in1k) |
| `regnety_008.pycls_in1k` | 224 | 76.334 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_008.pycls_in1k) |
| `regnety_016.pycls_in1k` | 224 | 77.880 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_016.pycls_in1k) |
| `regnety_032.pycls_in1k` | 224 | 78.902 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_032.pycls_in1k) |
| `regnety_040.pycls_in1k` | 224 | 79.246 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_040.pycls_in1k) |
| `regnety_064.pycls_in1k` | 224 | 79.734 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_064.pycls_in1k) |
| `regnety_080.pycls_in1k` | 224 | 79.874 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_080.pycls_in1k) |
| `regnety_120.pycls_in1k` | 224 | 80.382 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_120.pycls_in1k) |
| `regnety_160.pycls_in1k` | 224 | 80.284 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_160.pycls_in1k) |
| `regnety_320.pycls_in1k` | 224 | 80.796 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/regnety_320.pycls_in1k) |

### pytorch/vision

Origin: [github.com/pytorch/vision](https://github.com/pytorch/vision)

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `regnetx_004_tv.tv2_in1k` | 224 | 74.596 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnetx_004_tv.tv2_in1k) |
| `regnetx_008.tv2_in1k` | 224 | 77.294 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnetx_008.tv2_in1k) |
| `regnetx_016.tv2_in1k` | 224 | 79.440 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnetx_016.tv2_in1k) |
| `regnetx_032.tv2_in1k` | 224 | 80.916 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnetx_032.tv2_in1k) |
| `regnetx_080.tv2_in1k` | 224 | 81.552 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnetx_080.tv2_in1k) |
| `regnetx_160.tv2_in1k` | 224 | 82.574 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnetx_160.tv2_in1k) |
| `regnetx_320.tv2_in1k` | 224 | 82.824 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnetx_320.tv2_in1k) |
| `regnety_004.tv2_in1k` | 224 | 75.594 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnety_004.tv2_in1k) |
| `regnety_008_tv.tv2_in1k` | 224 | 78.672 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnety_008_tv.tv2_in1k) |
| `regnety_016.tv2_in1k` | 224 | 80.668 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnety_016.tv2_in1k) |
| `regnety_032.tv2_in1k` | 224 | 81.788 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnety_032.tv2_in1k) |
| `regnety_080_tv.tv2_in1k` | 224 | 82.580 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnety_080_tv.tv2_in1k) |
| `regnety_160.swag_ft_in1k` | 384 | 86.020 | 384 | 1000 | cc-by-nc-4.0 | [timm](https://huggingface.co/timm/regnety_160.swag_ft_in1k) |
| `regnety_160.swag_lc_in1k` | 224 | 83.806 | 224 | 1000 | cc-by-nc-4.0 | [timm](https://huggingface.co/timm/regnety_160.swag_lc_in1k) |
| `regnety_160.tv2_in1k` | 224 | 82.646 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnety_160.tv2_in1k) |
| `regnety_320.swag_ft_in1k` | 384 | 86.830 | 384 | 1000 | cc-by-nc-4.0 | [timm](https://huggingface.co/timm/regnety_320.swag_ft_in1k) |
| `regnety_320.swag_lc_in1k` | 224 | 84.654 | 224 | 1000 | cc-by-nc-4.0 | [timm](https://huggingface.co/timm/regnety_320.swag_lc_in1k) |
| `regnety_320.tv2_in1k` | 224 | 83.170 | 224 | 1000 | bsd-3-clause | [timm](https://huggingface.co/timm/regnety_320.tv2_in1k) |
| `regnety_1280.swag_ft_in1k` | 384 | 88.228 | 384 | 1000 | cc-by-nc-4.0 | [timm](https://huggingface.co/timm/regnety_1280.swag_ft_in1k) |
| `regnety_1280.swag_lc_in1k` | 224 | 85.964 | 224 | 1000 | cc-by-nc-4.0 | [timm](https://huggingface.co/timm/regnety_1280.swag_lc_in1k) |

### `ra3` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `regnetv_040.ra3_in1k` | 224 (test 288) | 83.204 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetv_040.ra3_in1k) |
| `regnetv_064.ra3_in1k` | 224 (test 288) | 83.746 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetv_064.ra3_in1k) |
| `regnety_040.ra3_in1k` | 224 (test 288) | 83.048 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnety_040.ra3_in1k) |
| `regnety_064.ra3_in1k` | 224 (test 288) | 83.716 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnety_064.ra3_in1k) |
| `regnety_080.ra3_in1k` | 224 (test 288) | 83.940 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnety_080.ra3_in1k) |
| `regnetz_040.ra3_in1k` | 256 (test 320) | 84.242 | 320 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetz_040.ra3_in1k) |
| `regnetz_040_h.ra3_in1k` | 256 (test 320) | 84.498 | 320 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetz_040_h.ra3_in1k) |

### facebookresearch/vissl

Origin: [github.com/facebookresearch/vissl](https://github.com/facebookresearch/vissl)

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `regnety_320.seer` | 224 | - | - | none | seer-license | [timm](https://huggingface.co/timm/regnety_320.seer) |
| `regnety_320.seer_ft_in1k` | 384 | 83.324 | 384 | 1000 | seer-license | [timm](https://huggingface.co/timm/regnety_320.seer_ft_in1k) |
| `regnety_640.seer` | 224 | - | - | none | seer-license | [timm](https://huggingface.co/timm/regnety_640.seer) |
| `regnety_640.seer_ft_in1k` | 384 | 83.924 | 384 | 1000 | seer-license | [timm](https://huggingface.co/timm/regnety_640.seer_ft_in1k) |
| `regnety_1280.seer` | 224 | - | - | none | seer-license | [timm](https://huggingface.co/timm/regnety_1280.seer) |
| `regnety_1280.seer_ft_in1k` | 384 | 84.426 | 384 | 1000 | seer-license | [timm](https://huggingface.co/timm/regnety_1280.seer_ft_in1k) |
| `regnety_2560.seer_ft_in1k` | 384 | 85.144 | 384 | 1000 | seer-license | [timm](https://huggingface.co/timm/regnety_2560.seer_ft_in1k) |

### `sw` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `regnety_120.sw_in12k` | 224 (test 288) | - | - | 11821 | apache-2.0 | [timm](https://huggingface.co/timm/regnety_120.sw_in12k) |
| `regnety_120.sw_in12k_ft_in1k` | 224 (test 288) | 85.428 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnety_120.sw_in12k_ft_in1k) |
| `regnety_160.sw_in12k` | 224 (test 288) | - | - | 11821 | apache-2.0 | [timm](https://huggingface.co/timm/regnety_160.sw_in12k) |
| `regnety_160.sw_in12k_ft_in1k` | 224 (test 288) | 86.028 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnety_160.sw_in12k_ft_in1k) |

### Other

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `regnety_032.ra_in1k` | 224 (test 288) | 82.724 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnety_032.ra_in1k) |
| `regnety_160.deit_in1k` | 224 (test 288) | 83.704 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnety_160.deit_in1k) |
| `regnety_160.lion_in12k_ft_in1k` | 224 (test 288) | 86.004 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnety_160.lion_in12k_ft_in1k) |