Model card for regnetz_c16.ra3_in1k
A RegNetZ image classification model. Trained on ImageNet-1k by Ross Wightman in timm
.
These RegNetZ B / C / D models explore different group size and layer configurations and did not follow any paper descriptions. Like EfficientNets, this architecture uses linear (non activated) block outputs and an inverted-bottleneck (mid block expansion).
- B16 : ~1.5GF @ 256x256 with a group-width of 16. Single layer stem.
- C16 : ~2.5GF @ 256x256 with a group-width of 16. Single layer stem.
- D32 : ~6GF @ 256x256 with a group-width of 32. Tiered 3-layer stem, no pooling.
- D8 : ~4GF @ 256x256 with a group-width of 8. Tiered 3-layer stem, no pooling.
- E8 : ~10GF @ 256x256 with a group-width of 8. Tiered 3-layer stem, no pooling.
This model architecture is implemented using timm
's flexible BYOBNet (Bring-Your-Own-Blocks Network).
BYOBNet allows configuration of:
- block / stage layout
- stem layout
- output stride (dilation)
- activation and norm layers
- channel and spatial / self-attention layers
...and also includes timm
features common to many other architectures, including:
- stochastic depth
- gradient checkpointing
- layer-wise LR decay
- per-stage feature extraction
Model Details
- Model Type: Image classification / feature backbone
- Model Stats:
- Params (M): 13.5
- GMACs: 2.5
- Activations (M): 16.6
- Image size: train = 256 x 256, test = 320 x 320
- Papers:
- Fast and Accurate Model Scaling: https://arxiv.org/abs/2103.06877
- Dataset: ImageNet-1k
- Original: https://github.com/huggingface/pytorch-image-models
Model Usage
Image Classification
from urllib.request import urlopen
from PIL import Image
import timm
img = Image.open(urlopen(
'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))
model = timm.create_model('regnetz_c16.ra3_in1k', pretrained=True)
model = model.eval()
# get model specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)
output = model(transforms(img).unsqueeze(0)) # unsqueeze single image into batch of 1
top5_probabilities, top5_class_indices = torch.topk(output.softmax(dim=1) * 100, k=5)
Feature Map Extraction
from urllib.request import urlopen
from PIL import Image
import timm
img = Image.open(urlopen(
'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))
model = timm.create_model(
'regnetz_c16.ra3_in1k',
pretrained=True,
features_only=True,
)
model = model.eval()
# get model specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)
output = model(transforms(img).unsqueeze(0)) # unsqueeze single image into batch of 1
for o in output:
# print shape of each feature map in output
# e.g.:
# torch.Size([1, 32, 128, 128])
# torch.Size([1, 48, 64, 64])
# torch.Size([1, 96, 32, 32])
# torch.Size([1, 192, 16, 16])
# torch.Size([1, 1536, 8, 8])
print(o.shape)
Image Embeddings
from urllib.request import urlopen
from PIL import Image
import timm
img = Image.open(urlopen(
'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))
model = timm.create_model(
'regnetz_c16.ra3_in1k',
pretrained=True,
num_classes=0, # remove classifier nn.Linear
)
model = model.eval()
# get model specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)
output = model(transforms(img).unsqueeze(0)) # output is (batch_size, num_features) shaped tensor
# or equivalently (without needing to set num_classes=0)
output = model.forward_features(transforms(img).unsqueeze(0))
# output is unpooled, a (1, 1536, 8, 8) shaped tensor
output = model.forward_head(output, pre_logits=True)
# output is a (1, num_features) shaped tensor
Model Comparison
Explore the dataset and runtime metrics of this model in timm model results.
For the comparison summary below, the ra_in1k, ra3_in1k, ch_in1k, sw_*, and lion_* tagged weights are trained in timm
.
model | img_size | top1 | top5 | param_count | gmacs | macts |
---|---|---|---|---|---|---|
regnety_1280.swag_ft_in1k | 384 | 88.228 | 98.684 | 644.81 | 374.99 | 210.2 |
regnety_320.swag_ft_in1k | 384 | 86.84 | 98.364 | 145.05 | 95.0 | 88.87 |
regnety_160.swag_ft_in1k | 384 | 86.024 | 98.05 | 83.59 | 46.87 | 67.67 |
regnety_160.sw_in12k_ft_in1k | 288 | 86.004 | 97.83 | 83.59 | 26.37 | 38.07 |
regnety_1280.swag_lc_in1k | 224 | 85.996 | 97.848 | 644.81 | 127.66 | 71.58 |
regnety_160.lion_in12k_ft_in1k | 288 | 85.982 | 97.844 | 83.59 | 26.37 | 38.07 |
regnety_160.sw_in12k_ft_in1k | 224 | 85.574 | 97.666 | 83.59 | 15.96 | 23.04 |
regnety_160.lion_in12k_ft_in1k | 224 | 85.564 | 97.674 | 83.59 | 15.96 | 23.04 |
regnety_120.sw_in12k_ft_in1k | 288 | 85.398 | 97.584 | 51.82 | 20.06 | 35.34 |
regnety_2560.seer_ft_in1k | 384 | 85.15 | 97.436 | 1282.6 | 747.83 | 296.49 |
regnetz_e8.ra3_in1k | 320 | 85.036 | 97.268 | 57.7 | 15.46 | 63.94 |
regnety_120.sw_in12k_ft_in1k | 224 | 84.976 | 97.416 | 51.82 | 12.14 | 21.38 |
regnety_320.swag_lc_in1k | 224 | 84.56 | 97.446 | 145.05 | 32.34 | 30.26 |
regnetz_040_h.ra3_in1k | 320 | 84.496 | 97.004 | 28.94 | 6.43 | 37.94 |
regnetz_e8.ra3_in1k | 256 | 84.436 | 97.02 | 57.7 | 9.91 | 40.94 |
regnety_1280.seer_ft_in1k | 384 | 84.432 | 97.092 | 644.81 | 374.99 | 210.2 |
regnetz_040.ra3_in1k | 320 | 84.246 | 96.93 | 27.12 | 6.35 | 37.78 |
regnetz_d8.ra3_in1k | 320 | 84.054 | 96.992 | 23.37 | 6.19 | 37.08 |
regnetz_d8_evos.ch_in1k | 320 | 84.038 | 96.992 | 23.46 | 7.03 | 38.92 |
regnetz_d32.ra3_in1k | 320 | 84.022 | 96.866 | 27.58 | 9.33 | 37.08 |
regnety_080.ra3_in1k | 288 | 83.932 | 96.888 | 39.18 | 13.22 | 29.69 |
regnety_640.seer_ft_in1k | 384 | 83.912 | 96.924 | 281.38 | 188.47 | 124.83 |
regnety_160.swag_lc_in1k | 224 | 83.778 | 97.286 | 83.59 | 15.96 | 23.04 |
regnetz_040_h.ra3_in1k | 256 | 83.776 | 96.704 | 28.94 | 4.12 | 24.29 |
regnetv_064.ra3_in1k | 288 | 83.72 | 96.75 | 30.58 | 10.55 | 27.11 |
regnety_064.ra3_in1k | 288 | 83.718 | 96.724 | 30.58 | 10.56 | 27.11 |
regnety_160.deit_in1k | 288 | 83.69 | 96.778 | 83.59 | 26.37 | 38.07 |
regnetz_040.ra3_in1k | 256 | 83.62 | 96.704 | 27.12 | 4.06 | 24.19 |
regnetz_d8.ra3_in1k | 256 | 83.438 | 96.776 | 23.37 | 3.97 | 23.74 |
regnetz_d32.ra3_in1k | 256 | 83.424 | 96.632 | 27.58 | 5.98 | 23.74 |
regnetz_d8_evos.ch_in1k | 256 | 83.36 | 96.636 | 23.46 | 4.5 | 24.92 |
regnety_320.seer_ft_in1k | 384 | 83.35 | 96.71 | 145.05 | 95.0 | 88.87 |
regnetv_040.ra3_in1k | 288 | 83.204 | 96.66 | 20.64 | 6.6 | 20.3 |
regnety_320.tv2_in1k | 224 | 83.162 | 96.42 | 145.05 | 32.34 | 30.26 |
regnety_080.ra3_in1k | 224 | 83.16 | 96.486 | 39.18 | 8.0 | 17.97 |
regnetv_064.ra3_in1k | 224 | 83.108 | 96.458 | 30.58 | 6.39 | 16.41 |
regnety_040.ra3_in1k | 288 | 83.044 | 96.5 | 20.65 | 6.61 | 20.3 |
regnety_064.ra3_in1k | 224 | 83.02 | 96.292 | 30.58 | 6.39 | 16.41 |
regnety_160.deit_in1k | 224 | 82.974 | 96.502 | 83.59 | 15.96 | 23.04 |
regnetx_320.tv2_in1k | 224 | 82.816 | 96.208 | 107.81 | 31.81 | 36.3 |
regnety_032.ra_in1k | 288 | 82.742 | 96.418 | 19.44 | 5.29 | 18.61 |
regnety_160.tv2_in1k | 224 | 82.634 | 96.22 | 83.59 | 15.96 | 23.04 |
regnetz_c16_evos.ch_in1k | 320 | 82.634 | 96.472 | 13.49 | 3.86 | 25.88 |
regnety_080_tv.tv2_in1k | 224 | 82.592 | 96.246 | 39.38 | 8.51 | 19.73 |
regnetx_160.tv2_in1k | 224 | 82.564 | 96.052 | 54.28 | 15.99 | 25.52 |
regnetz_c16.ra3_in1k | 320 | 82.51 | 96.358 | 13.46 | 3.92 | 25.88 |
regnetv_040.ra3_in1k | 224 | 82.44 | 96.198 | 20.64 | 4.0 | 12.29 |
regnety_040.ra3_in1k | 224 | 82.304 | 96.078 | 20.65 | 4.0 | 12.29 |
regnetz_c16.ra3_in1k | 256 | 82.16 | 96.048 | 13.46 | 2.51 | 16.57 |
regnetz_c16_evos.ch_in1k | 256 | 81.936 | 96.15 | 13.49 | 2.48 | 16.57 |
regnety_032.ra_in1k | 224 | 81.924 | 95.988 | 19.44 | 3.2 | 11.26 |
regnety_032.tv2_in1k | 224 | 81.77 | 95.842 | 19.44 | 3.2 | 11.26 |
regnetx_080.tv2_in1k | 224 | 81.552 | 95.544 | 39.57 | 8.02 | 14.06 |
regnetx_032.tv2_in1k | 224 | 80.924 | 95.27 | 15.3 | 3.2 | 11.37 |
regnety_320.pycls_in1k | 224 | 80.804 | 95.246 | 145.05 | 32.34 | 30.26 |
regnetz_b16.ra3_in1k | 288 | 80.712 | 95.47 | 9.72 | 2.39 | 16.43 |
regnety_016.tv2_in1k | 224 | 80.66 | 95.334 | 11.2 | 1.63 | 8.04 |
regnety_120.pycls_in1k | 224 | 80.37 | 95.12 | 51.82 | 12.14 | 21.38 |
regnety_160.pycls_in1k | 224 | 80.288 | 94.964 | 83.59 | 15.96 | 23.04 |
regnetx_320.pycls_in1k | 224 | 80.246 | 95.01 | 107.81 | 31.81 | 36.3 |
regnety_080.pycls_in1k | 224 | 79.882 | 94.834 | 39.18 | 8.0 | 17.97 |
regnetz_b16.ra3_in1k | 224 | 79.872 | 94.974 | 9.72 | 1.45 | 9.95 |
regnetx_160.pycls_in1k | 224 | 79.862 | 94.828 | 54.28 | 15.99 | 25.52 |
regnety_064.pycls_in1k | 224 | 79.716 | 94.772 | 30.58 | 6.39 | 16.41 |
regnetx_120.pycls_in1k | 224 | 79.592 | 94.738 | 46.11 | 12.13 | 21.37 |
regnetx_016.tv2_in1k | 224 | 79.44 | 94.772 | 9.19 | 1.62 | 7.93 |
regnety_040.pycls_in1k | 224 | 79.23 | 94.654 | 20.65 | 4.0 | 12.29 |
regnetx_080.pycls_in1k | 224 | 79.198 | 94.55 | 39.57 | 8.02 | 14.06 |
regnetx_064.pycls_in1k | 224 | 79.064 | 94.454 | 26.21 | 6.49 | 16.37 |
regnety_032.pycls_in1k | 224 | 78.884 | 94.412 | 19.44 | 3.2 | 11.26 |
regnety_008_tv.tv2_in1k | 224 | 78.654 | 94.388 | 6.43 | 0.84 | 5.42 |
regnetx_040.pycls_in1k | 224 | 78.482 | 94.24 | 22.12 | 3.99 | 12.2 |
regnetx_032.pycls_in1k | 224 | 78.178 | 94.08 | 15.3 | 3.2 | 11.37 |
regnety_016.pycls_in1k | 224 | 77.862 | 93.73 | 11.2 | 1.63 | 8.04 |
regnetx_008.tv2_in1k | 224 | 77.302 | 93.672 | 7.26 | 0.81 | 5.15 |
regnetx_016.pycls_in1k | 224 | 76.908 | 93.418 | 9.19 | 1.62 | 7.93 |
regnety_008.pycls_in1k | 224 | 76.296 | 93.05 | 6.26 | 0.81 | 5.25 |
regnety_004.tv2_in1k | 224 | 75.592 | 92.712 | 4.34 | 0.41 | 3.89 |
regnety_006.pycls_in1k | 224 | 75.244 | 92.518 | 6.06 | 0.61 | 4.33 |
regnetx_008.pycls_in1k | 224 | 75.042 | 92.342 | 7.26 | 0.81 | 5.15 |
regnetx_004_tv.tv2_in1k | 224 | 74.57 | 92.184 | 5.5 | 0.42 | 3.17 |
regnety_004.pycls_in1k | 224 | 74.018 | 91.764 | 4.34 | 0.41 | 3.89 |
regnetx_006.pycls_in1k | 224 | 73.862 | 91.67 | 6.2 | 0.61 | 3.98 |
regnetx_004.pycls_in1k | 224 | 72.38 | 90.832 | 5.16 | 0.4 | 3.14 |
regnety_002.pycls_in1k | 224 | 70.282 | 89.534 | 3.16 | 0.2 | 2.17 |
regnetx_002.pycls_in1k | 224 | 68.752 | 88.556 | 2.68 | 0.2 | 2.16 |
Citation
@misc{rw2019timm,
author = {Ross Wightman},
title = {PyTorch Image Models},
year = {2019},
publisher = {GitHub},
journal = {GitHub repository},
doi = {10.5281/zenodo.4414861},
howpublished = {\url{https://github.com/huggingface/pytorch-image-models}}
}
@InProceedings{Dollar2021,
title = {Fast and Accurate Model Scaling},
author = {Piotr Doll{'a}r and Mannat Singh and Ross Girshick},
booktitle = {CVPR},
year = {2021}
}
- Downloads last month
- 6,721