# MetaFormer (PoolFormer, ConvFormer, CAFormer)

**Category:** Hybrid · **Code:**
[`timm/models/metaformer.py`](https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/metaformer.py) ·
**Architectures:** 18 · **Pretrained weights:** 50 (42 with ImageNet-1k results)

MetaFormer models share a general transformer block structure and vary the token mixer: pooling in PoolFormer /
PoolFormerV2, separable convolutions in ConvFormer, and convolutions in the early stages with attention in the later
stages in CAFormer.

## Papers

* [MetaFormer Is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418)
* [MetaFormer Baselines for Vision](https://arxiv.org/abs/2210.13452)

## 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('caformer_s18.sail_in1k', pretrained=True)
>>> data_config = timm.data.resolve_model_data_config(model)
```

## Model API support

| Feature | `MetaFormer` |
|---|---|
| [`features_only=True` feature maps](../model_api#intermediate-features) | ✓ (strides 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 (`MetaFormer` with `poolformerv2_s12` 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 |
|---|---:|---:|---:|---:|
| `caformer_b36` | 98.75 | 23.22 | 224 | 5 |
| `caformer_m36` | 56.20 | 13.29 | 224 | 5 |
| `caformer_s18` | 26.34 | 4.13 | 224 | 5 |
| `caformer_s36` | 39.30 | 8.00 | 224 | 5 |
| `convformer_b36` | 99.88 | 22.69 | 224 | 5 |
| `convformer_m36` | 57.05 | 12.89 | 224 | 5 |
| `convformer_s18` | 26.77 | 3.96 | 224 | 5 |
| `convformer_s36` | 40.01 | 7.67 | 224 | 5 |
| `poolformer_m36` | 56.17 | 8.80 | 224 | 1 |
| `poolformer_m48` | 73.47 | 11.59 | 224 | 1 |
| `poolformer_s12` | 11.92 | 1.82 | 224 | 1 |
| `poolformer_s24` | 21.39 | 3.41 | 224 | 1 |
| `poolformer_s36` | 30.86 | 5.00 | 224 | 1 |
| `poolformerv2_m36` | 56.08 | 8.81 | 224 | 1 |
| `poolformerv2_m48` | 73.35 | 11.59 | 224 | 1 |
| `poolformerv2_s12` | 11.89 | 1.83 | 224 | 1 |
| `poolformerv2_s24` | 21.34 | 3.42 | 224 | 1 |
| `poolformerv2_s36` | 30.79 | 5.01 | 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.

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `caformer_b36.sail_in1k` | 224 | 85.502 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_b36.sail_in1k) |
| `caformer_b36.sail_in1k_384` | 384 | 86.400 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_b36.sail_in1k_384) |
| `caformer_b36.sail_in22k` | 224 | - | - | 21841 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_b36.sail_in22k) |
| `caformer_b36.sail_in22k_ft_in1k` | 224 | 87.450 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_b36.sail_in22k_ft_in1k) |
| `caformer_b36.sail_in22k_ft_in1k_384` | 384 | 88.058 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_b36.sail_in22k_ft_in1k_384) |
| `caformer_m36.sail_in1k` | 224 | 85.242 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_m36.sail_in1k) |
| `caformer_m36.sail_in1k_384` | 384 | 86.164 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_m36.sail_in1k_384) |
| `caformer_m36.sail_in22k` | 224 | - | - | 21841 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_m36.sail_in22k) |
| `caformer_m36.sail_in22k_ft_in1k` | 224 | 86.606 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_m36.sail_in22k_ft_in1k) |
| `caformer_m36.sail_in22k_ft_in1k_384` | 384 | 87.470 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_m36.sail_in22k_ft_in1k_384) |
| `caformer_s18.sail_in1k` | 224 | 83.664 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_s18.sail_in1k) |
| `caformer_s18.sail_in1k_384` | 384 | 85.036 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_s18.sail_in1k_384) |
| `caformer_s18.sail_in22k` | 224 | - | - | 21841 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_s18.sail_in22k) |
| `caformer_s18.sail_in22k_ft_in1k` | 224 | 84.092 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_s18.sail_in22k_ft_in1k) |
| `caformer_s18.sail_in22k_ft_in1k_384` | 384 | 85.442 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_s18.sail_in22k_ft_in1k_384) |
| `caformer_s36.sail_in1k` | 224 | 84.528 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_s36.sail_in1k) |
| `caformer_s36.sail_in1k_384` | 384 | 85.762 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_s36.sail_in1k_384) |
| `caformer_s36.sail_in22k` | 224 | - | - | 21841 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_s36.sail_in22k) |
| `caformer_s36.sail_in22k_ft_in1k` | 224 | 85.780 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_s36.sail_in22k_ft_in1k) |
| `caformer_s36.sail_in22k_ft_in1k_384` | 384 | 86.860 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/caformer_s36.sail_in22k_ft_in1k_384) |
| `convformer_b36.sail_in1k` | 224 | 84.826 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_b36.sail_in1k) |
| `convformer_b36.sail_in1k_384` | 384 | 85.770 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_b36.sail_in1k_384) |
| `convformer_b36.sail_in22k` | 224 | - | - | 21841 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_b36.sail_in22k) |
| `convformer_b36.sail_in22k_ft_in1k` | 224 | 86.992 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_b36.sail_in22k_ft_in1k) |
| `convformer_b36.sail_in22k_ft_in1k_384` | 384 | 87.606 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_b36.sail_in22k_ft_in1k_384) |
| `convformer_m36.sail_in1k` | 224 | 84.486 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_m36.sail_in1k) |
| `convformer_m36.sail_in1k_384` | 384 | 85.580 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_m36.sail_in1k_384) |
| `convformer_m36.sail_in22k` | 224 | - | - | 21841 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_m36.sail_in22k) |
| `convformer_m36.sail_in22k_ft_in1k` | 224 | 86.152 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_m36.sail_in22k_ft_in1k) |
| `convformer_m36.sail_in22k_ft_in1k_384` | 384 | 86.868 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_m36.sail_in22k_ft_in1k_384) |
| `convformer_s18.sail_in1k` | 224 | 82.968 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_s18.sail_in1k) |
| `convformer_s18.sail_in1k_384` | 384 | 84.392 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_s18.sail_in1k_384) |
| `convformer_s18.sail_in22k` | 224 | - | - | 21841 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_s18.sail_in22k) |
| `convformer_s18.sail_in22k_ft_in1k` | 224 | 83.768 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_s18.sail_in22k_ft_in1k) |
| `convformer_s18.sail_in22k_ft_in1k_384` | 384 | 85.004 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_s18.sail_in22k_ft_in1k_384) |
| `convformer_s36.sail_in1k` | 224 | 84.068 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_s36.sail_in1k) |
| `convformer_s36.sail_in1k_384` | 384 | 85.380 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_s36.sail_in1k_384) |
| `convformer_s36.sail_in22k` | 224 | - | - | 21841 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_s36.sail_in22k) |
| `convformer_s36.sail_in22k_ft_in1k` | 224 | 85.420 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_s36.sail_in22k_ft_in1k) |
| `convformer_s36.sail_in22k_ft_in1k_384` | 384 | 86.374 | 384 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/convformer_s36.sail_in22k_ft_in1k_384) |
| `poolformer_m36.sail_in1k` | 224 | 82.104 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/poolformer_m36.sail_in1k) |
| `poolformer_m48.sail_in1k` | 224 | 82.462 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/poolformer_m48.sail_in1k) |
| `poolformer_s12.sail_in1k` | 224 | 77.238 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/poolformer_s12.sail_in1k) |
| `poolformer_s24.sail_in1k` | 224 | 80.294 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/poolformer_s24.sail_in1k) |
| `poolformer_s36.sail_in1k` | 224 | 81.428 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/poolformer_s36.sail_in1k) |
| `poolformerv2_m36.sail_in1k` | 224 | 82.214 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/poolformerv2_m36.sail_in1k) |
| `poolformerv2_m48.sail_in1k` | 224 | 82.612 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/poolformerv2_m48.sail_in1k) |
| `poolformerv2_s12.sail_in1k` | 224 | 78.008 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/poolformerv2_s12.sail_in1k) |
| `poolformerv2_s24.sail_in1k` | 224 | 80.746 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/poolformerv2_s24.sail_in1k) |
| `poolformerv2_s36.sail_in1k` | 224 | 81.576 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/poolformerv2_s36.sail_in1k) |

