fiftyone.utils.clip.model.ModifiedResNet

class documentation

class ModifiedResNet(nn.Module): (source)

Constructor: ModifiedResNet(layers, output_dim, heads, input_resolution, width)

View In Hierarchy

A ResNet class that is similar to torchvision's but contains the following changes:

  • There are now 3 "stem" convolutions as opposed to 1, with an average pool instead of a max pool.
  • Performs anti-aliasing strided convolutions, where an avgpool is prepended to convolutions with stride > 1
  • The final pooling layer is a QKV attention instead of an average pool
Method __init__ Undocumented
Method forward Undocumented
Instance Variable attnpool Undocumented
Instance Variable avgpool Undocumented
Instance Variable bn1 Undocumented
Instance Variable bn2 Undocumented
Instance Variable bn3 Undocumented
Instance Variable conv1 Undocumented
Instance Variable conv2 Undocumented
Instance Variable conv3 Undocumented
Instance Variable input_resolution Undocumented
Instance Variable layer1 Undocumented
Instance Variable layer2 Undocumented
Instance Variable layer3 Undocumented
Instance Variable layer4 Undocumented
Instance Variable output_dim Undocumented
Instance Variable relu Undocumented
Method _make_layer Undocumented
Instance Variable _inplanes Undocumented
def __init__(self, layers, output_dim, heads, input_resolution=224, width=64): (source)

Undocumented

def forward(self, x): (source)

Undocumented

attnpool = (source)

Undocumented

avgpool = (source)

Undocumented

bn1 = (source)

Undocumented

bn2 = (source)

Undocumented

bn3 = (source)

Undocumented

conv1 = (source)

Undocumented

conv2 = (source)

Undocumented

conv3 = (source)

Undocumented

input_resolution = (source)

Undocumented

layer1 = (source)

Undocumented

layer2 = (source)

Undocumented

layer3 = (source)

Undocumented

layer4 = (source)

Undocumented

output_dim = (source)

Undocumented

relu = (source)

Undocumented

def _make_layer(self, planes, blocks, stride=1): (source)

Undocumented

_inplanes = (source)

Undocumented