# FasterNet

**Category:** Efficient/Mobile · **Code:**
[`timm/models/fasternet.py`](https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/fasternet.py) ·
**Architectures:** 6 · **Pretrained weights:** 6 (6 with ImageNet-1k results)

FasterNet is built around partial convolution (PConv), which convolves only a part of the input channels to reduce
redundant computation and memory access.

## Papers

* [Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks](https://arxiv.org/abs/2303.03667)

## 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('fasternet_s.in1k', pretrained=True)
>>> data_config = timm.data.resolve_model_data_config(model)
```

## Model API support

| Feature | `FasterNet` |
|---|---|
| [`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 (`FasterNet` with `fasternet_t0` 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 |
|---|---:|---:|---:|---:|
| `fasternet_l` | 93.47 | 15.52 | 224 | 1 |
| `fasternet_m` | 53.52 | 8.74 | 224 | 1 |
| `fasternet_s` | 31.18 | 4.56 | 224 | 1 |
| `fasternet_t0` | 3.91 | 0.34 | 224 | 1 |
| `fasternet_t1` | 7.60 | 0.85 | 224 | 1 |
| `fasternet_t2` | 14.98 | 1.91 | 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 |
|---|---:|---:|---:|---:|---|---|
| `fasternet_l.in1k` | 224 | 83.516 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/fasternet_l.in1k) |
| `fasternet_m.in1k` | 224 | 82.848 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/fasternet_m.in1k) |
| `fasternet_s.in1k` | 224 | 81.328 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/fasternet_s.in1k) |
| `fasternet_t0.in1k` | 224 | 71.732 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/fasternet_t0.in1k) |
| `fasternet_t1.in1k` | 224 | 75.922 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/fasternet_t1.in1k) |
| `fasternet_t2.in1k` | 224 | 78.738 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/fasternet_t2.in1k) |

## Citation

```bibtex
@article{chen2023run,
  title={Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks},
  author={Chen, Jierun and Kao, Shiu-hong and He, Hao and Zhuo, Weipeng and Wen, Song and Lee, Chul-Ho and Chan, S-H Gary},
  journal={arXiv preprint arXiv:2303.03667},
  year={2023}
}
```