--- tags: - image-classification - timm library_name: timm license: apache-2.0 datasets: - imagenet-1k --- # Model card for resnet50d.ra4_e3600_r224_in1k A ResNet image classification model. Trained on ImageNet-1k by Ross Wightman. Trained with `timm` scripts using hyper-parameters inspired by the MobileNet-V4 small, mixed with go-to hparams from `timm` and "ResNet Strikes Back". A collection of hparams (timm .yaml config files) for this training series can be found here: https://gist.github.com/rwightman/f6705cb65c03daeebca8aa129b1b94ad ## Model Details - **Model Type:** Image classification / feature backbone - **Model Stats:** - Params (M): 25.6 - GMACs: 4.4 - Activations (M): 11.9 - Image size: train = 224 x 224, test = 288 x 288 - **Dataset:** ImageNet-1k - **Papers:** - PyTorch Image Models: https://github.com/huggingface/pytorch-image-models - ResNet strikes back: An improved training procedure in timm: https://arxiv.org/abs/2110.00476 - Deep Residual Learning for Image Recognition: https://arxiv.org/abs/1512.03385 - MobileNetV4 -- Universal Models for the Mobile Ecosystem: https://arxiv.org/abs/2404.10518 ## Model Usage ### Image Classification ```python 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('resnet50d.ra4_e3600_r224_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 ```python 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( 'resnet50d.ra4_e3600_r224_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, 64, 112, 112]) # torch.Size([1, 256, 56, 56]) # torch.Size([1, 512, 28, 28]) # torch.Size([1, 1024, 14, 14]) # torch.Size([1, 2048, 7, 7]) print(o.shape) ``` ### Image Embeddings ```python 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( 'resnet50d.ra4_e3600_r224_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, 2048, 7, 7) 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](http://hf.co/timm/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](http://hf.co/timm/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](http://hf.co/timm/mobilenetv4_conv_aa_large.e230_r448_in12k_ft_in1k) | 84.64 | 97.114 | 32.59 | 448 | | [mobilenetv4_hybrid_large.ix_e600_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_large.ix_e600_r384_in1k) | 84.356 | 96.892 | 37.76 | 448 | | [mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k](http://hf.co/timm/mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k) | 84.314 | 97.102 | 32.59 | 384 | | [mobilenetv4_hybrid_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_large.e600_r384_in1k) | 84.266 | 96.936 | 37.76 | 448 | | [mobilenetv4_hybrid_large.ix_e600_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_large.ix_e600_r384_in1k) | 83.990 | 96.702 | 37.76 | 384 | | [mobilenetv4_conv_aa_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_conv_aa_large.e600_r384_in1k) | 83.824 | 96.734 | 32.59 | 480 | | [mobilenetv4_hybrid_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_large.e600_r384_in1k) | 83.800 | 96.770 | 37.76 | 384 | | [mobilenetv4_hybrid_medium.ix_e550_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.ix_e550_r384_in1k) | 83.394 | 96.760 | 11.07 | 448 | | [mobilenetv4_conv_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_conv_large.e600_r384_in1k) | 83.392 | 96.622 | 32.59 | 448 | | [mobilenetv4_conv_aa_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_conv_aa_large.e600_r384_in1k) | 83.244 | 96.392 | 32.59 | 384 | | [mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k) | 82.99 | 96.67 | 11.07 | 320 | | [mobilenetv4_hybrid_medium.ix_e550_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.ix_e550_r384_in1k) | 82.968 | 96.474 | 11.07 | 384 | | [mobilenetv4_conv_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_conv_large.e600_r384_in1k) | 82.952 | 96.266 | 32.59 | 384 | | [mobilenetv4_conv_large.e500_r256_in1k](http://hf.co/timm/mobilenetv4_conv_large.e500_r256_in1k) | 82.674 | 96.31 | 32.59 | 320 | | [mobilenetv4_hybrid_medium.ix_e550_r256_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.ix_e550_r256_in1k) | 82.492 | 96.278 | 11.07 | 320 | | [mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k) | 82.364 | 96.256 | 11.07 | 256 | | [mobilenetv4_conv_large.e500_r256_in1k](http://hf.co/timm/mobilenetv4_conv_large.e500_r256_in1k) | 81.862 | 95.69 | 32.59 | 256 | | [resnet50d.ra4_e3600_r224_in1k](http://hf.co/timm/resnet50d.ra4_e3600_r224_in1k) | 81.838 | 95.922 | 25.58 | 288 | | [mobilenetv3_large_150d.ra4_e3600_r256_in1k](http://hf.co/timm/mobilenetv3_large_150d.ra4_e3600_r256_in1k) | 81.806 | 95.9 | 14.62 | 320 | | [mobilenetv4_hybrid_medium.ix_e550_r256_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.ix_e550_r256_in1k) | 81.446 | 95.704 | 11.07 | 256 | | [efficientnet_b1.ra4_e3600_r240_in1k](http://hf.co/timm/efficientnet_b1.ra4_e3600_r240_in1k) | 81.440 | 95.700 | 7.79 | 288 | | [mobilenetv4_hybrid_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.e500_r224_in1k) | 81.276 | 95.742 | 11.07 | 256 | | [resnet50d.ra4_e3600_r224_in1k](http://hf.co/timm/resnet50d.ra4_e3600_r224_in1k) | 80.952 | 95.384 | 25.58 | 224 | | [mobilenetv3_large_150d.ra4_e3600_r256_in1k](http://hf.co/timm/mobilenetv3_large_150d.ra4_e3600_r256_in1k) | 80.944 | 95.448 | 14.62 | 256 | | [mobilenetv4_conv_medium.e500_r256_in1k](http://hf.co/timm/mobilenetv4_conv_medium.e500_r256_in1k) | 80.858 | 95.768 | 9.72 | 320 | | [mobilenet_edgetpu_v2_m.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenet_edgetpu_v2_m.ra4_e3600_r224_in1k) | 80.680 | 95.442 | 8.46 | 256 | | [mobilenetv4_hybrid_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.e500_r224_in1k) | 80.442 | 95.38 | 11.07 | 224 | | [efficientnet_b1.ra4_e3600_r240_in1k](http://hf.co/timm/efficientnet_b1.ra4_e3600_r240_in1k) | 80.406 | 95.152 | 7.79 | 240 | | [mobilenetv4_conv_blur_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_conv_blur_medium.e500_r224_in1k) | 80.142 | 95.298 | 9.72 | 256 | | [mobilenet_edgetpu_v2_m.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenet_edgetpu_v2_m.ra4_e3600_r224_in1k) | 80.130 | 95.002 | 8.46 | 224 | | [mobilenetv4_conv_medium.e500_r256_in1k](http://hf.co/timm/mobilenetv4_conv_medium.e500_r256_in1k) | 79.928 | 95.184 | 9.72 | 256 | | [mobilenetv4_conv_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_conv_medium.e500_r224_in1k) | 79.808 | 95.186 | 9.72 | 256 | | [resnetv2_34d.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_34d.ra4_e3600_r224_in1k) | 79.590 | 94.770 | 21.82 | 288 | | [mobilenetv4_conv_blur_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_conv_blur_medium.e500_r224_in1k) | 79.438 | 94.932 | 9.72 | 224 | | [efficientnet_b0.ra4_e3600_r224_in1k](http://hf.co/timm/efficientnet_b0.ra4_e3600_r224_in1k) | 79.364 | 94.754 | 5.29 | 256 | | [mobilenetv4_conv_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_conv_medium.e500_r224_in1k) | 79.094 | 94.77 | 9.72 | 224 | | [resnetv2_34.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_34.ra4_e3600_r224_in1k) | 79.072 | 94.566 | 21.80 | 288 | | [resnet34.ra4_e3600_r224_in1k](http://hf.co/timm/resnet34.ra4_e3600_r224_in1k) | 78.952 | 94.450 | 21.80 | 288 | | [efficientnet_b0.ra4_e3600_r224_in1k](http://hf.co/timm/efficientnet_b0.ra4_e3600_r224_in1k) | 78.584 | 94.338 | 5.29 | 224 | | [resnetv2_34d.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_34d.ra4_e3600_r224_in1k) | 78.268 | 93.952 | 21.82 | 224 | | [resnetv2_34.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_34.ra4_e3600_r224_in1k) | 77.636 | 93.528 | 21.80 | 224 | | [mobilenetv1_125.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_125.ra4_e3600_r224_in1k) | 77.600 | 93.804 | 6.27 | 256 | | [resnet34.ra4_e3600_r224_in1k](http://hf.co/timm/resnet34.ra4_e3600_r224_in1k) | 77.448 | 93.502 | 21.80 | 224 | | [mobilenetv3_large_100.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv3_large_100.ra4_e3600_r224_in1k) | 77.164 | 93.336 | 5.48 | 256 | | [mobilenetv1_125.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_125.ra4_e3600_r224_in1k) | 76.924 | 93.234 | 6.27 | 224 | | [mobilenetv1_100h.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_100h.ra4_e3600_r224_in1k) | 76.596 | 93.272 | 5.28 | 256 | | [mobilenetv3_large_100.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv3_large_100.ra4_e3600_r224_in1k) | 76.310 | 92.846 | 5.48 | 224 | | [mobilenetv1_100.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_100.ra4_e3600_r224_in1k) | 76.094 | 93.004 | 4.23 | 256 | | [resnetv2_18d.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_18d.ra4_e3600_r224_in1k) | 76.044 | 93.020 | 11.71 | 288 | | [resnet18d.ra4_e3600_r224_in1k](http://hf.co/timm/resnet18d.ra4_e3600_r224_in1k) | 76.024 | 92.780 | 11.71 | 288 | | [mobilenetv1_100h.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_100h.ra4_e3600_r224_in1k) | 75.662 | 92.504 | 5.28 | 224 | | [mobilenetv1_100.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_100.ra4_e3600_r224_in1k) | 75.382 | 92.312 | 4.23 | 224 | | [resnetv2_18.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_18.ra4_e3600_r224_in1k) | 75.340 | 92.678 | 11.69 | 288 | | [mobilenetv4_conv_small.e2400_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small.e2400_r224_in1k) | 74.616 | 92.072 | 3.77 | 256 | | [resnetv2_18d.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_18d.ra4_e3600_r224_in1k) | 74.412 | 91.936 | 11.71 | 224 | | [resnet18d.ra4_e3600_r224_in1k](http://hf.co/timm/resnet18d.ra4_e3600_r224_in1k) | 74.322 | 91.832 | 11.71 | 224 | | [mobilenetv4_conv_small.e1200_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small.e1200_r224_in1k) | 74.292 | 92.116 | 3.77 | 256 | | [mobilenetv4_conv_small.e2400_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small.e2400_r224_in1k) | 73.756 | 91.422 | 3.77 | 224 | | [resnetv2_18.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_18.ra4_e3600_r224_in1k) | 73.578 | 91.352 | 11.69 | 224 | | [mobilenetv4_conv_small.e1200_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small.e1200_r224_in1k) | 73.454 | 91.34 | 3.77 | 224 | | [mobilenetv4_conv_small_050.e3000_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small_050.e3000_r224_in1k) | 65.810 | 86.424 | 2.24 | 256 | | [mobilenetv4_conv_small_050.e3000_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small_050.e3000_r224_in1k) | 64.762 | 85.514 | 2.24 | 224 | ## Citation ```bibtex @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}} } ``` ```bibtex @inproceedings{wightman2021resnet, title={ResNet strikes back: An improved training procedure in timm}, author={Wightman, Ross and Touvron, Hugo and Jegou, Herve}, booktitle={NeurIPS 2021 Workshop on ImageNet: Past, Present, and Future} } ``` ```bibtex @article{He2015, author = {Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun}, title = {Deep Residual Learning for Image Recognition}, journal = {arXiv preprint arXiv:1512.03385}, year = {2015} } ``` ```bibtex @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} } ```