MobileNetV4 s

Click a block to read its description, or select it with Tab and Enter.

MobileNetV4 sClassification, 224 × 224 input, 1,000 classes. Tensor sizes exclude batch.LibreYOLOMobileNetV4 sClassification, 224 × 224 input, 1,000 classes. Tensor sizes exclude batch.NetworkInput3 × 224 × 224Conv2d 3×3 / 232 × 112 × 112; p=1, g=1BatchNorm2d32 × 112 × 112ReLU32 × 112 × 112Layer 1: ConvBnAct32 × 56 × 56, repeat 1Layer 1: ConvBnAct32 × 56 × 56, repeat 1Layer 2: ConvBnAct96 × 28 × 28, repeat 1Layer 2: ConvBnAct64 × 28 × 28, repeat 1Layer 3: UniversalInvertedResidual96 × 14 × 14, repeat 1Layer 3: UniversalInvertedResidual96 × 14 × 14, repeat 4Layer 3: UniversalInvertedResidual96 × 14 × 14, repeat 1Layer 4: UniversalInvertedResidual128 × 7 × 7, repeat 1Layer 4: UniversalInvertedResidual128 × 7 × 7, repeat 1Layer 4: UniversalInvertedResidual128 × 7 × 7, repeat 1Layer 4: UniversalInvertedResidual128 × 7 × 7, repeat 1Layer 4: UniversalInvertedResidual128 × 7 × 7, repeat 2Layer 5: ConvBnAct960 × 7 × 7, repeat 1AdaptiveAvgPool2d960 × 1 × 1Conv2d 1×1 / 11280 × 1 × 1; p=0, g=1BatchNorm2d1280 × 1 × 1ReLU1,280 × 1 × 1Flatten1,280Linear1280 to 1000Layer 1, unit 1ConvBnActInput 32 × 112 × 112Conv2d 3×3 / 232 × 56 × 56; p=1, g=1BatchNorm2d32 × 56 × 56ReLU32 × 56 × 56Output 32 × 56 × 56Repeat 1 timesLayer 1, unit 2ConvBnActInput 32 × 56 × 56Conv2d 1×1 / 132 × 56 × 56; p=0, g=1BatchNorm2d32 × 56 × 56ReLU32 × 56 × 56Output 32 × 56 × 56Repeat 1 timesLayer 2, unit 1ConvBnActInput 32 × 56 × 56Conv2d 3×3 / 296 × 28 × 28; p=1, g=1BatchNorm2d96 × 28 × 28ReLU96 × 28 × 28Output 96 × 28 × 28Repeat 1 timesLayer 2, unit 2ConvBnActInput 96 × 28 × 28Conv2d 1×1 / 164 × 28 × 28; p=0, g=1BatchNorm2d64 × 28 × 28ReLU64 × 28 × 28Output 64 × 28 × 28Repeat 1 timesLayer 3, unit 1UniversalInvertedResidualInput 64 × 28 × 28Conv2d 5×5 / 164 × 28 × 28; p=2, g=64BatchNorm2d64 × 28 × 28Conv2d 1×1 / 1192 × 28 × 28; p=0, g=1BatchNorm2d192 × 28 × 28ReLU192 × 28 × 28Conv2d 5×5 / 2192 × 14 × 14; p=2, g=192BatchNorm2d192 × 14 × 14ReLU192 × 14 × 14Conv2d 1×1 / 196 × 14 × 14; p=0, g=1BatchNorm2d96 × 14 × 14Output 96 × 14 × 14Repeat 1 timesLayer 3, unit 2UniversalInvertedResidualInput 96 × 14 × 14Conv2d 1×1 / 1192 × 14 × 14; p=0, g=1BatchNorm2d192 × 14 × 14ReLU192 × 14 × 14Conv2d 3×3 / 1192 × 14 × 14; p=1, g=192BatchNorm2d192 × 14 × 14ReLU192 × 14 × 14Conv2d 1×1 / 196 × 14 × 14; p=0, g=1BatchNorm2d96 × 14 × 14+Output 96 × 14 × 14Repeat 4 timesLayer 3, unit 6UniversalInvertedResidualInput 96 × 14 × 14Conv2d 3×3 / 196 × 14 × 14; p=1, g=96BatchNorm2d96 × 14 × 14Conv2d 1×1 / 1384 × 14 × 14; p=0, g=1BatchNorm2d384 × 14 × 14ReLU384 × 14 × 14Conv2d 1×1 / 196 × 14 × 14; p=0, g=1BatchNorm2d96 × 14 × 14+Output 96 × 14 × 14Repeat 1 timesLayer 4, unit 1UniversalInvertedResidualInput 96 × 14 × 14Conv2d 3×3 / 196 × 14 × 14; p=1, g=96BatchNorm2d96 × 14 × 14Conv2d 1×1 / 1576 × 14 × 14; p=0, g=1BatchNorm2d576 × 14 × 14ReLU576 × 14 × 14Conv2d 3×3 / 2576 × 7 × 7; p=1, g=576BatchNorm2d576 × 7 × 7ReLU576 × 7 × 7Conv2d 1×1 / 1128 × 7 × 7; p=0, g=1BatchNorm2d128 × 7 × 7Output 128 × 7 × 7Repeat 1 timesLayer 4, unit 2UniversalInvertedResidualInput 128 × 7 × 7Conv2d 5×5 / 1128 × 7 × 7; p=2, g=128BatchNorm2d128 × 7 × 7Conv2d 1×1 / 1512 × 7 × 7; p=0, g=1BatchNorm2d512 × 7 × 7ReLU512 × 7 × 7Conv2d 5×5 / 1512 × 7 × 7; p=2, g=512BatchNorm2d512 × 7 × 7ReLU512 × 7 × 7Conv2d 1×1 / 1128 × 7 × 7; p=0, g=1BatchNorm2d128 × 7 × 7+Output 128 × 7 × 7Repeat 1 timesLayer 4, unit 3UniversalInvertedResidualInput 128 × 7 × 7Conv2d 1×1 / 1512 × 7 × 7; p=0, g=1BatchNorm2d512 × 7 × 7ReLU512 × 7 × 7Conv2d 5×5 / 1512 × 7 × 7; p=2, g=512BatchNorm2d512 × 7 × 7ReLU512 × 7 × 7Conv2d 1×1 / 1128 × 7 × 7; p=0, g=1BatchNorm2d128 × 7 × 7+Output 128 × 7 × 7Repeat 1 timesLayer 4, unit 4UniversalInvertedResidualInput 128 × 7 × 7Conv2d 1×1 / 1384 × 7 × 7; p=0, g=1BatchNorm2d384 × 7 × 7ReLU384 × 7 × 7Conv2d 5×5 / 1384 × 7 × 7; p=2, g=384BatchNorm2d384 × 7 × 7ReLU384 × 7 × 7Conv2d 1×1 / 1128 × 7 × 7; p=0, g=1BatchNorm2d128 × 7 × 7+Output 128 × 7 × 7Repeat 1 timesLayer 4, unit 5UniversalInvertedResidualInput 128 × 7 × 7Conv2d 1×1 / 1512 × 7 × 7; p=0, g=1BatchNorm2d512 × 7 × 7ReLU512 × 7 × 7Conv2d 3×3 / 1512 × 7 × 7; p=1, g=512BatchNorm2d512 × 7 × 7ReLU512 × 7 × 7Conv2d 1×1 / 1128 × 7 × 7; p=0, g=1BatchNorm2d128 × 7 × 7+Output 128 × 7 × 7Repeat 2 timesLayer 5, unit 1ConvBnActInput 128 × 7 × 7Conv2d 1×1 / 1960 × 7 × 7; p=0, g=1BatchNorm2d960 × 7 × 7ReLU960 × 7 × 7Output 960 × 7 × 7Repeat 1 timesRepeated units are grouped only when operation parameters, tensor shapes and shortcut behavior match.Source: libreyolo/models/mobilenetv4/nn.py. Revision a4d0ecc9e17f.libreyolo.com