# BYOBNet (GENet, RepVGG, MobileOne, ...)

**Category:** CNN · **Code:**
[`timm/models/byobnet.py`](https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/byobnet.py) ·
**Architectures:** 53 · **Pretrained weights:** 57 (41 with ImageNet-1k results)

Bring-Your-Own-Blocks Network, a flexible network built from dataclass configs that stack blocks into stages. It
implements GPU-Efficient Networks (`gernet_*`), RepVGG, MobileOne, RegNetZ, the ResNet image towers of CLIP (`*_clip`)
and a number of ResNet style `timm` variants.

## Papers

* [Neural Architecture Design for GPU-Efficient Networks](https://arxiv.org/abs/2006.14090)
* [RepVGG: Making VGG-style ConvNets Great Again](https://arxiv.org/abs/2101.03697)
* [MobileOne: An Improved One millisecond Mobile Backbone](https://arxiv.org/abs/2206.04040)
* [Fast and Accurate Model Scaling](https://arxiv.org/abs/2103.06877)
* [Evolving Normalization-Activation Layers](https://arxiv.org/abs/2004.02967)
* [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020)

## 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('repvgg_a2.rvgg_in1k', pretrained=True)
>>> data_config = timm.data.resolve_model_data_config(model)
```

## Model API support

| Feature | `ByobNet` |
|---|---|
| [`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 (`ByobNet` with `mobileone_s1` 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 |
|---|---:|---:|---:|---:|
| `bat_resnext26ts` | 10.73 | 2.53 | 256 | 1 |
| `eca_resnet33ts` | 19.68 | 4.76 | 256 | 1 |
| `eca_resnext26ts` | 10.30 | 2.43 | 256 | 1 |
| `gcresnet33ts` | 19.88 | 4.76 | 256 | 1 |
| `gcresnet50t` | 25.90 | 5.42 | 256 | 1 |
| `gcresnext26ts` | 10.48 | 2.43 | 256 | 1 |
| `gcresnext50ts` | 15.67 | 3.75 | 256 | 1 |
| `gernet_l` | 31.08 | 4.57 | 256 | 1 |
| `gernet_m` | 21.14 | 3.02 | 224 | 1 |
| `gernet_s` | 8.17 | 0.75 | 224 | 1 |
| `mobileone_s0` | 2.08 | 0.28 | 224 | 1 |
| `mobileone_s1` | 4.76 | 0.83 | 224 | 1 |
| `mobileone_s2` | 7.81 | 1.30 | 224 | 1 |
| `mobileone_s3` | 10.08 | 1.90 | 224 | 1 |
| `mobileone_s4` | 14.84 | 2.98 | 224 | 1 |
| `regnetz_b16` | 9.72 | 1.45 | 224 | 1 |
| `regnetz_b16_evos` | 9.74 | 1.43 | 224 | 0 |
| `regnetz_c16` | 13.46 | 2.51 | 256 | 1 |
| `regnetz_c16_evos` | 13.49 | 2.48 | 256 | 1 |
| `regnetz_d8` | 23.37 | 3.97 | 256 | 1 |
| `regnetz_d8_evos` | 23.46 | 4.50 | 256 | 1 |
| `regnetz_d32` | 27.58 | 5.98 | 256 | 1 |
| `regnetz_e8` | 57.70 | 9.91 | 256 | 1 |
| `repvgg_a0` | 8.31 | 1.36 | 224 | 1 |
| `repvgg_a1` | 12.79 | 2.36 | 224 | 1 |
| `repvgg_a2` | 25.50 | 5.12 | 224 | 1 |
| `repvgg_b0` | 14.34 | 3.06 | 224 | 1 |
| `repvgg_b1` | 51.83 | 11.82 | 224 | 1 |
| `repvgg_b1g4` | 36.13 | 7.31 | 224 | 1 |
| `repvgg_b2` | 80.32 | 18.38 | 224 | 1 |
| `repvgg_b2g4` | 55.78 | 11.33 | 224 | 1 |
| `repvgg_b3` | 110.96 | 26.21 | 224 | 1 |
| `repvgg_b3g4` | 75.63 | 16.06 | 224 | 1 |
| `repvgg_d2se` | 120.39 | 66.99 | 320 | 1 |
| `resnet32ts` | 17.96 | 4.63 | 256 | 1 |
| `resnet33ts` | 19.68 | 4.76 | 256 | 1 |
| `resnet50_clip` | 38.32 | 6.14 | 224 | 3 |
| `resnet50_clip_gap` | 23.53 | 5.39 | 224 | 3 |
| `resnet50_mlp` | 26.65 | 7.05 | 256 | 0 |
| `resnet50x4_clip` | 87.14 | 21.35 | 288 | 1 |
| `resnet50x4_clip_gap` | 65.62 | 19.57 | 288 | 1 |
| `resnet50x16_clip` | 167.33 | 74.90 | 384 | 1 |
| `resnet50x16_clip_gap` | 136.20 | 70.32 | 384 | 1 |
| `resnet50x64_clip` | 420.38 | 265.02 | 448 | 1 |
| `resnet50x64_clip_gap` | 365.03 | 253.96 | 448 | 1 |
| `resnet51q` | 35.70 | 6.38 | 256 | 1 |
| `resnet61q` | 36.85 | 7.80 | 256 | 1 |
| `resnet101_clip` | 56.26 | 9.81 | 224 | 2 |
| `resnet101_clip_gap` | 42.52 | 9.11 | 224 | 2 |
| `resnext26ts` | 10.30 | 2.43 | 256 | 1 |
| `seresnet33ts` | 19.78 | 4.76 | 256 | 1 |
| `seresnext26ts` | 10.39 | 2.43 | 256 | 1 |
| `test_byobnet` | 0.46 | 0.03 | 160 | 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.

### `rvgg` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `repvgg_a0.rvgg_in1k` | 224 | 72.426 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_a0.rvgg_in1k) |
| `repvgg_a1.rvgg_in1k` | 224 | 74.474 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_a1.rvgg_in1k) |
| `repvgg_a2.rvgg_in1k` | 224 | 76.488 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_a2.rvgg_in1k) |
| `repvgg_b0.rvgg_in1k` | 224 | 75.158 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_b0.rvgg_in1k) |
| `repvgg_b1.rvgg_in1k` | 224 | 78.370 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_b1.rvgg_in1k) |
| `repvgg_b1g4.rvgg_in1k` | 224 | 77.608 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_b1g4.rvgg_in1k) |
| `repvgg_b2.rvgg_in1k` | 224 | 78.788 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_b2.rvgg_in1k) |
| `repvgg_b2g4.rvgg_in1k` | 224 | 79.396 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_b2g4.rvgg_in1k) |
| `repvgg_b3.rvgg_in1k` | 224 | 80.528 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_b3.rvgg_in1k) |
| `repvgg_b3g4.rvgg_in1k` | 224 | 80.224 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_b3g4.rvgg_in1k) |
| `repvgg_d2se.rvgg_in1k` | 320 | 83.550 | 320 | 1000 | mit | [timm](https://huggingface.co/timm/repvgg_d2se.rvgg_in1k) |

### `openai` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `resnet50_clip.openai` | 224 | - | - | 1024 | mit | [timm](https://huggingface.co/timm/resnet50_clip.openai) |
| `resnet50_clip_gap.openai` | 224 | - | - | none | mit | [timm](https://huggingface.co/timm/resnet50_clip_gap.openai) |
| `resnet50x4_clip.openai` | 288 | - | - | 640 | mit | [timm](https://huggingface.co/timm/resnet50x4_clip.openai) |
| `resnet50x4_clip_gap.openai` | 288 | - | - | none | mit | [timm](https://huggingface.co/timm/resnet50x4_clip_gap.openai) |
| `resnet50x16_clip.openai` | 384 | - | - | 768 | mit | [timm](https://huggingface.co/timm/resnet50x16_clip.openai) |
| `resnet50x16_clip_gap.openai` | 384 | - | - | none | mit | [timm](https://huggingface.co/timm/resnet50x16_clip_gap.openai) |
| `resnet50x64_clip.openai` | 448 | - | - | 1024 | mit | [timm](https://huggingface.co/timm/resnet50x64_clip.openai) |
| `resnet50x64_clip_gap.openai` | 448 | - | - | none | mit | [timm](https://huggingface.co/timm/resnet50x64_clip_gap.openai) |
| `resnet101_clip.openai` | 224 | - | - | 512 | mit | [timm](https://huggingface.co/timm/resnet101_clip.openai) |
| `resnet101_clip_gap.openai` | 224 | - | - | none | mit | [timm](https://huggingface.co/timm/resnet101_clip_gap.openai) |

### `ra2` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `eca_resnet33ts.ra2_in1k` | 256 (test 288) | 80.716 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/eca_resnet33ts.ra2_in1k) |
| `gcresnet33ts.ra2_in1k` | 256 (test 288) | 80.588 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/gcresnet33ts.ra2_in1k) |
| `gcresnet50t.ra2_in1k` | 256 (test 288) | 81.458 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/gcresnet50t.ra2_in1k) |
| `resnet32ts.ra2_in1k` | 256 (test 288) | 79.388 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnet32ts.ra2_in1k) |
| `resnet33ts.ra2_in1k` | 256 (test 288) | 79.710 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnet33ts.ra2_in1k) |
| `resnet51q.ra2_in1k` | 256 (test 288) | 82.356 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnet51q.ra2_in1k) |
| `resnet61q.ra2_in1k` | 256 (test 288) | 82.532 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnet61q.ra2_in1k) |
| `resnext26ts.ra2_in1k` | 256 (test 288) | 77.186 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/resnext26ts.ra2_in1k) |
| `seresnet33ts.ra2_in1k` | 256 (test 288) | 80.814 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/seresnet33ts.ra2_in1k) |

### `ch` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `bat_resnext26ts.ch_in1k` | 256 | 78.280 | 256 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/bat_resnext26ts.ch_in1k) |
| `eca_resnext26ts.ch_in1k` | 256 (test 288) | 78.020 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/eca_resnext26ts.ch_in1k) |
| `gcresnext26ts.ch_in1k` | 256 (test 288) | 78.436 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/gcresnext26ts.ch_in1k) |
| `gcresnext50ts.ch_in1k` | 256 (test 288) | 81.242 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/gcresnext50ts.ch_in1k) |
| `regnetz_c16_evos.ch_in1k` | 256 (test 320) | 82.660 | 320 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetz_c16_evos.ch_in1k) |
| `regnetz_d8_evos.ch_in1k` | 256 (test 320) | 84.138 | 320 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetz_d8_evos.ch_in1k) |
| `seresnext26ts.ch_in1k` | 256 (test 288) | 78.292 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/seresnext26ts.ch_in1k) |

### `apple` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `mobileone_s0.apple_in1k` | 224 | 71.398 | 224 | 1000 | mobileone-license | [timm](https://huggingface.co/timm/mobileone_s0.apple_in1k) |
| `mobileone_s1.apple_in1k` | 224 | 75.760 | 224 | 1000 | mobileone-license | [timm](https://huggingface.co/timm/mobileone_s1.apple_in1k) |
| `mobileone_s2.apple_in1k` | 224 | 77.504 | 224 | 1000 | mobileone-license | [timm](https://huggingface.co/timm/mobileone_s2.apple_in1k) |
| `mobileone_s3.apple_in1k` | 224 | 78.002 | 224 | 1000 | mobileone-license | [timm](https://huggingface.co/timm/mobileone_s3.apple_in1k) |
| `mobileone_s4.apple_in1k` | 224 | 79.446 | 224 | 1000 | mobileone-license | [timm](https://huggingface.co/timm/mobileone_s4.apple_in1k) |

### `ra3` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `regnetz_b16.ra3_in1k` | 224 (test 288) | 80.740 | 288 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetz_b16.ra3_in1k) |
| `regnetz_c16.ra3_in1k` | 256 (test 320) | 82.662 | 320 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetz_c16.ra3_in1k) |
| `regnetz_d8.ra3_in1k` | 256 (test 320) | 84.044 | 320 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetz_d8.ra3_in1k) |
| `regnetz_d32.ra3_in1k` | 256 (test 320) | 84.028 | 320 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetz_d32.ra3_in1k) |
| `regnetz_e8.ra3_in1k` | 256 (test 320) | 85.016 | 320 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/regnetz_e8.ra3_in1k) |

### `yfcc15m` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `resnet50_clip.yfcc15m` | 224 | - | - | 1024 | mit | [timm](https://huggingface.co/timm/resnet50_clip.yfcc15m) |
| `resnet50_clip_gap.yfcc15m` | 224 | - | - | none | mit | [timm](https://huggingface.co/timm/resnet50_clip_gap.yfcc15m) |
| `resnet101_clip.yfcc15m` | 224 | - | - | 512 | mit | [timm](https://huggingface.co/timm/resnet101_clip.yfcc15m) |
| `resnet101_clip_gap.yfcc15m` | 224 | - | - | none | mit | [timm](https://huggingface.co/timm/resnet101_clip_gap.yfcc15m) |

### `idstcv` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `gernet_l.idstcv_in1k` | 256 | 81.354 | 256 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/gernet_l.idstcv_in1k) |
| `gernet_m.idstcv_in1k` | 224 | 80.702 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/gernet_m.idstcv_in1k) |
| `gernet_s.idstcv_in1k` | 224 | 76.874 | 224 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/gernet_s.idstcv_in1k) |

### `cc12m` tags

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `resnet50_clip.cc12m` | 224 | - | - | 1024 | mit | [timm](https://huggingface.co/timm/resnet50_clip.cc12m) |
| `resnet50_clip_gap.cc12m` | 224 | - | - | none | mit | [timm](https://huggingface.co/timm/resnet50_clip_gap.cc12m) |

### Other

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `test_byobnet.r160_in1k` | 160 | 45.858 | 160 | 1000 | apache-2.0 | [timm](https://huggingface.co/timm/test_byobnet.r160_in1k) |