pytorch-pretrained-gluonresnet

A stand-alone version of the pretrained MxNet Gluon ResNet models ported to Pytorch. Currently part of my model collection (https://github.com/rwightman/pytorch-image-models/blob/master/models/gluon_resnet.py)

This model covers all variants of ResNet, ResNeXt, SE-ResNeXt, and SENet found in the gluon model zoo (https://gluon-cv.mxnet.io/model_zoo/classification.html, https://github.com/dmlc/gluon-cv/blob/master/gluoncv/model_zoo/model_zoo.py) that

• have stride in 3x3 conv layer of bottleneck
• have conv-bn-act ordering

Included ResNet variants are:

• v1b - 7x7 stem, stem_width=64, same as torchvision ResNet (checkpoint compatible), or NVIDIA ResNet 'v1.5'
• v1c - 3 layer deep 3x3 stem, stem_width = 32
• v1d - 3 layer deep 3x3 stem, stem_width = 32, average pool in downsample
• v1e - 3 layer deep 3x3 stem, stem_width = 64, average pool in downsample *no pretrained weights available
• v1s - 3 layer deep 3x3 stem, stem_width = 64

ResNeXt is standard and checkpoint compatible with torchvision pretrained models. 7x7 stem, stem_width = 64, standard cardinality and base width calcs

SE-ResNeXt is standard. 7x7 stem, stem_width = 64, checkpoints are not compatible with Cadene pretrained, but could be with key mapping

SENet-154 is standard. 3 layer deep 3x3 stem (same as v1c-v1s), stem_width = 64, cardinality=64, reduction by 2 on width of first bottleneck convolution, 3x3 downsample convs after first block

Original ResNet-V1, ResNet-V2 (bn-act-conv), and SE-ResNet (stride in first bottleneck conv) are NOT supported.

Results

PyTorch Hub

Models can be access via the PyTorch Hub API

>>> torch.hub.list('rwightman/pytorch-pretrained-gluonresnet')
['gluon_resnet18_v1b', ...]
>>> model = torch.hub.load('rwightman/pytorch-pretrained-gluonresnet', 'gluon_resnet50_v1d', pretrained=True)
>>> model.eval()
>>> output = model(torch.randn(1,3,224,224))