# ResNetV2 (BiT)

**Category:** CNN · **Code:**
[`timm/models/resnetv2.py`](https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/resnetv2.py) ·
**Architectures:** 20 · **Pretrained weights:** 24 (18 with ImageNet-1k results)

Pre-activation ResNet (v2) with Group Normalization and Weight Standardization as used by Big Transfer (BiT), with the
original BiT weights. Also supports the non pre-activation variant used as the hybrid Vision Transformer backbone.

## Papers

* [Big Transfer (BiT): General Visual Representation Learning](https://arxiv.org/abs/1912.11370)
* [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929)
* [Knowledge distillation: A good teacher is patient and consistent](https://arxiv.org/abs/2106.05237)
* [Identity Mappings in Deep Residual Networks](https://arxiv.org/abs/1603.05027)

## 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('resnetv2_50.a1h_in1k', pretrained=True)
>>> data_config = timm.data.resolve_model_data_config(model)
```

## Model API support

| Feature | `ResNetV2` |
|---|---|
| [`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 (`ResNetV2` with `resnetv2_18` 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 |
|---|---:|---:|---:|---:|
| `resnetv2_18` | 11.69 | 1.82 | 224 | 1 |
| `resnetv2_18d` | 11.71 | 2.06 | 224 | 1 |
| `resnetv2_34` | 21.80 | 3.67 | 224 | 1 |
| `resnetv2_34d` | 21.82 | 3.91 | 224 | 2 |
| `resnetv2_50` | 25.55 | 4.11 | 224 | 1 |
| `resnetv2_50d` | 25.57 | 4.35 | 224 | 0 |
| `resnetv2_50d_evos` | 25.59 | 4.33 | 224 | 1 |
| `resnetv2_50d_frn` | 25.59 | 4.33 | 224 | 0 |
| `resnetv2_50d_gn` | 25.57 | 4.38 | 224 | 1 |
| `resnetv2_50t` | 25.57 | 4.32 | 224 | 0 |
| `resnetv2_50x1_bit` | 25.55 | 4.23 | 224 | 3 |
| `resnetv2_50x3_bit` | 217.32 | 145.70 | 448 | 2 |
| `resnetv2_101` | 44.54 | 7.83 | 224 | 1 |
| `resnetv2_101d` | 44.56 | 8.07 | 224 | 0 |
| `resnetv2_101x1_bit` | 44.54 | 31.65 | 448 | 2 |
| `resnetv2_101x3_bit` | 387.93 | 280.33 | 448 | 2 |
| `resnetv2_152` | 60.19 | 11.55 | 224 | 0 |
| `resnetv2_152d` | 60.20 | 11.80 | 224 | 0 |
| `resnetv2_152x2_bit` | 236.34 | 46.95 | 224 | 4 |
| `resnetv2_152x4_bit` | 936.53 | 844.84 | 480 | 2 |

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.

### `goog` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `resnetv2_50x1_bit.goog_distilled_in1k` | 224 | 82.820 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_50x1_bit.goog_distilled_in1k) |
| `resnetv2_50x1_bit.goog_in21k` | 224 | - | - | 21843 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_50x1_bit.goog_in21k) |
| `resnetv2_50x1_bit.goog_in21k_ft_in1k` | 448 | 80.386 | 448 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_50x1_bit.goog_in21k_ft_in1k) |
| `resnetv2_50x3_bit.goog_in21k` | 224 | - | - | 21843 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_50x3_bit.goog_in21k) |
| `resnetv2_50x3_bit.goog_in21k_ft_in1k` | 448 | 83.982 | 448 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_50x3_bit.goog_in21k_ft_in1k) |
| `resnetv2_101x1_bit.goog_in21k` | 224 | - | - | 21843 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_101x1_bit.goog_in21k) |
| `resnetv2_101x1_bit.goog_in21k_ft_in1k` | 448 | 82.332 | 448 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_101x1_bit.goog_in21k_ft_in1k) |
| `resnetv2_101x3_bit.goog_in21k` | 224 | - | - | 21843 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_101x3_bit.goog_in21k) |
| `resnetv2_101x3_bit.goog_in21k_ft_in1k` | 448 | 84.430 | 448 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_101x3_bit.goog_in21k_ft_in1k) |
| `resnetv2_152x2_bit.goog_in21k` | 224 | - | - | 21843 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_152x2_bit.goog_in21k) |
| `resnetv2_152x2_bit.goog_in21k_ft_in1k` | 448 | 84.468 | 448 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_152x2_bit.goog_in21k_ft_in1k) |
| `resnetv2_152x2_bit.goog_teacher_in21k_ft_in1k` | 224 | 82.918 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_152x2_bit.goog_teacher_in21k_ft_in1k) |
| `resnetv2_152x2_bit.goog_teacher_in21k_ft_in1k_384` | 384 | 83.858 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_152x2_bit.goog_teacher_in21k_ft_in1k_384) |
| `resnetv2_152x4_bit.goog_in21k` | 224 | - | - | 21843 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_152x4_bit.goog_in21k) |
| `resnetv2_152x4_bit.goog_in21k_ft_in1k` | 480 | 84.936 | 480 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_152x4_bit.goog_in21k_ft_in1k) |

### `ra4` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `resnetv2_18.ra4_e3600_r224_in1k` | 224 (test 288) | 75.364 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_18.ra4_e3600_r224_in1k) |
| `resnetv2_18d.ra4_e3600_r224_in1k` | 224 (test 288) | 76.058 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_18d.ra4_e3600_r224_in1k) |
| `resnetv2_34.ra4_e3600_r224_in1k` | 224 (test 288) | 79.070 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_34.ra4_e3600_r224_in1k) |
| `resnetv2_34d.ra4_e3600_r224_in1k` | 224 (test 288) | 79.614 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_34d.ra4_e3600_r224_in1k) |
| `resnetv2_34d.ra4_e3600_r384_in1k` | 384 (test 448) | 80.430 | 448 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_34d.ra4_e3600_r384_in1k) |

### `a1h` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `resnetv2_50.a1h_in1k` | 224 (test 288) | 81.406 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_50.a1h_in1k) |
| `resnetv2_101.a1h_in1k` | 224 (test 288) | 83.002 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_101.a1h_in1k) |

### `ah` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `resnetv2_50d_evos.ah_in1k` | 224 (test 288) | 82.024 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_50d_evos.ah_in1k) |
| `resnetv2_50d_gn.ah_in1k` | 224 (test 288) | 81.960 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnetv2_50d_gn.ah_in1k) |