Model card for mobilenetv4_hybrid_large.e600_r384_in1k

A MobileNet-V4 image classification model. Trained on ImageNet-1k by Ross Wightman.

Trained with timm scripts using hyper-parameters inspired by the MobileNet-V4 paper with timm enhancements.

NOTE: So far, these are the only known MNV4 weights. Official weights for Tensorflow models are unreleased.

Model Details

• Model Type: Image classification / feature backbone
• Model Stats:
  • Params (M): 37.8
  • GMACs: 7.8
  • Activations (M): 34.5
  • Image size: train = 384 x 384, test = 448 x 448
• Dataset: ImageNet-1k
• Original: https://github.com/tensorflow/models/tree/master/official/vision
• Papers:
  • MobileNetV4 -- Universal Models for the Mobile Ecosystem: https://arxiv.org/abs/2404.10518
  • PyTorch Image Models: 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('mobilenetv4_hybrid_large.e600_r384_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(
    'mobilenetv4_hybrid_large.e600_r384_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, 24, 192, 192])
    #  torch.Size([1, 48, 96, 96])
    #  torch.Size([1, 96, 48, 48])
    #  torch.Size([1, 192, 24, 24])
    #  torch.Size([1, 960, 12, 12])

    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(
    'mobilenetv4_hybrid_large.e600_r384_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, 960, 12, 12) shaped tensor

output = model.forward_head(output, pre_logits=True)
# output is a (1, num_features) shaped tensor

Model Comparison

By Top-1

model	top1	top5	param_count	img_size
mobilenetv4_conv_aa_large.e230_r448_in12k_ft_in1k	84.99	97.294	32.59	544
mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k	84.772	97.344	32.59	480
mobilenetv4_conv_aa_large.e230_r448_in12k_ft_in1k	84.64	97.114	32.59	448
mobilenetv4_hybrid_large.ix_e600_r384_in1k	84.356	96.892	37.76	448
mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k	84.314	97.102	32.59	384
mobilenetv4_hybrid_large.e600_r384_in1k	84.266	96.936	37.76	448
mobilenetv4_hybrid_large.ix_e600_r384_in1k	83.990	96.702	37.76	384
mobilenetv4_conv_aa_large.e600_r384_in1k	83.824	96.734	32.59	480
mobilenetv4_hybrid_large.e600_r384_in1k	83.800	96.770	37.76	384
mobilenetv4_hybrid_medium.ix_e550_r384_in1k	83.394	96.760	11.07	448
mobilenetv4_conv_large.e600_r384_in1k	83.392	96.622	32.59	448
mobilenetv4_conv_aa_large.e600_r384_in1k	83.244	96.392	32.59	384
mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k	82.99	96.67	11.07	320
mobilenetv4_hybrid_medium.ix_e550_r384_in1k	82.968	96.474	11.07	384
mobilenetv4_conv_large.e600_r384_in1k	82.952	96.266	32.59	384
mobilenetv4_conv_large.e500_r256_in1k	82.674	96.31	32.59	320
mobilenetv4_hybrid_medium.ix_e550_r256_in1k	82.492	96.278	11.07	320
mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k	82.364	96.256	11.07	256
mobilenetv4_conv_large.e500_r256_in1k	81.862	95.69	32.59	256
resnet50d.ra4_e3600_r224_in1k	81.838	95.922	25.58	288
mobilenetv3_large_150d.ra4_e3600_r256_in1k	81.806	95.9	14.62	320
mobilenetv4_hybrid_medium.ix_e550_r256_in1k	81.446	95.704	11.07	256
efficientnet_b1.ra4_e3600_r240_in1k	81.440	95.700	7.79	288
mobilenetv4_hybrid_medium.e500_r224_in1k	81.276	95.742	11.07	256
resnet50d.ra4_e3600_r224_in1k	80.952	95.384	25.58	224
mobilenetv3_large_150d.ra4_e3600_r256_in1k	80.944	95.448	14.62	256
mobilenetv4_conv_medium.e500_r256_in1k	80.858	95.768	9.72	320
mobilenet_edgetpu_v2_m.ra4_e3600_r224_in1k	80.680	95.442	8.46	256
mobilenetv4_hybrid_medium.e500_r224_in1k	80.442	95.38	11.07	224
efficientnet_b1.ra4_e3600_r240_in1k	80.406	95.152	7.79	240
mobilenetv4_conv_blur_medium.e500_r224_in1k	80.142	95.298	9.72	256
mobilenet_edgetpu_v2_m.ra4_e3600_r224_in1k	80.130	95.002	8.46	224
mobilenetv4_conv_medium.e500_r256_in1k	79.928	95.184	9.72	256
mobilenetv4_conv_medium.e500_r224_in1k	79.808	95.186	9.72	256
mobilenetv4_conv_blur_medium.e500_r224_in1k	79.438	94.932	9.72	224
efficientnet_b0.ra4_e3600_r224_in1k	79.364	94.754	5.29	256
mobilenetv4_conv_medium.e500_r224_in1k	79.094	94.77	9.72	224
efficientnet_b0.ra4_e3600_r224_in1k	78.584	94.338	5.29	224
mobilenetv1_125.ra4_e3600_r224_in1k	77.600	93.804	6.27	256
mobilenetv3_large_100.ra4_e3600_r224_in1k	77.164	93.336	5.48	256
mobilenetv1_125.ra4_e3600_r224_in1k	76.924	93.234	6.27	224
mobilenetv1_100h.ra4_e3600_r224_in1k	76.596	93.272	5.28	256
mobilenetv3_large_100.ra4_e3600_r224_in1k	76.310	92.846	5.48	224
mobilenetv1_100.ra4_e3600_r224_in1k	76.094	93.004	4.23	256
mobilenetv1_100h.ra4_e3600_r224_in1k	75.662	92.504	5.28	224
mobilenetv1_100.ra4_e3600_r224_in1k	75.382	92.312	4.23	224
mobilenetv4_conv_small.e2400_r224_in1k	74.616	92.072	3.77	256
mobilenetv4_conv_small.e1200_r224_in1k	74.292	92.116	3.77	256
mobilenetv4_conv_small.e2400_r224_in1k	73.756	91.422	3.77	224
mobilenetv4_conv_small.e1200_r224_in1k	73.454	91.34	3.77	224

Citation

@article{qin2024mobilenetv4,
  title={MobileNetV4-Universal Models for the Mobile Ecosystem},
  author={Qin, Danfeng and Leichner, Chas and Delakis, Manolis and Fornoni, Marco and Luo, Shixin and Yang, Fan and Wang, Weijun and Banbury, Colby and Ye, Chengxi and Akin, Berkin and others},
  journal={arXiv preprint arXiv:2404.10518},
  year={2024}
}

@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}}
}