mobilenet_v2

torchvision.models.quantization.mobilenet_v2(*, weights: Optional[Union[MobileNet_V2_QuantizedWeights, MobileNet_V2_Weights]] = None, progress: bool = True, quantize: bool = False, **kwargs: Any) → QuantizableMobileNetV2

Constructs a MobileNetV2 architecture from MobileNetV2: Inverted Residuals and Linear Bottlenecks.

Note

Note that quantize = True returns a quantized model with 8 bit weights. Quantized models only support inference and run on CPUs. GPU inference is not yet supported.

Parameters:

• weights (MobileNet_V2_QuantizedWeights or MobileNet_V2_Weights, optional) – The pretrained weights for the model. See MobileNet_V2_QuantizedWeights below for more details, and possible values. By default, no pre-trained weights are used.

• progress (bool, optional) – If True, displays a progress bar of the download to stderr. Default is True.

• quantize (bool, optional) – If True, returns a quantized version of the model. Default is False.

• **kwargs – parameters passed to the torchvision.models.quantization.QuantizableMobileNetV2 base class. Please refer to the source code for more details about this class.

class torchvision.models.quantization.MobileNet_V2_QuantizedWeights(value)

The model builder above accepts the following values as the weights parameter. MobileNet_V2_QuantizedWeights.DEFAULT is equivalent to MobileNet_V2_QuantizedWeights.IMAGENET1K_QNNPACK_V1. You can also use strings, e.g. weights='DEFAULT' or weights='IMAGENET1K_QNNPACK_V1'.

MobileNet_V2_QuantizedWeights.IMAGENET1K_QNNPACK_V1:

These weights were produced by doing Quantization Aware Training (eager mode) on top of the unquantized weights listed below. Also available as MobileNet_V2_QuantizedWeights.DEFAULT.

acc@1 (on ImageNet-1K)	71.658
acc@5 (on ImageNet-1K)	90.15
num_params	3504872
min_size	height=1, width=1
categories	tench, goldfish, great white shark, … (997 omitted)
backend	qnnpack
recipe	link
unquantized	MobileNet_V2_Weights.IMAGENET1K_V1

The inference transforms are available at MobileNet_V2_QuantizedWeights.IMAGENET1K_QNNPACK_V1.transforms and perform the following preprocessing operations: Accepts PIL.Image, batched (B, C, H, W) and single (C, H, W) image torch.Tensor objects. The images are resized to resize_size=[256] using interpolation=InterpolationMode.BILINEAR, followed by a central crop of crop_size=[224]. Finally the values are first rescaled to [0.0, 1.0] and then normalized using mean=[0.485, 0.456, 0.406] and std=[0.229, 0.224, 0.225].

class torchvision.models.MobileNet_V2_Weights(value)

The model builder above accepts the following values as the weights parameter. MobileNet_V2_Weights.DEFAULT is equivalent to MobileNet_V2_Weights.IMAGENET1K_V2. You can also use strings, e.g. weights='DEFAULT' or weights='IMAGENET1K_V1'.

MobileNet_V2_Weights.IMAGENET1K_V1:

These weights reproduce closely the results of the paper using a simple training recipe.

acc@1 (on ImageNet-1K)	71.878
acc@5 (on ImageNet-1K)	90.286
num_params	3504872
min_size	height=1, width=1
categories	tench, goldfish, great white shark, … (997 omitted)
recipe	link

The inference transforms are available at MobileNet_V2_Weights.IMAGENET1K_V1.transforms and perform the following preprocessing operations: Accepts PIL.Image, batched (B, C, H, W) and single (C, H, W) image torch.Tensor objects. The images are resized to resize_size=[256] using interpolation=InterpolationMode.BILINEAR, followed by a central crop of crop_size=[224]. Finally the values are first rescaled to [0.0, 1.0] and then normalized using mean=[0.485, 0.456, 0.406] and std=[0.229, 0.224, 0.225].

MobileNet_V2_Weights.IMAGENET1K_V2:

These weights improve upon the results of the original paper by using a modified version of TorchVision’s new training recipe. Also available as MobileNet_V2_Weights.DEFAULT.

acc@1 (on ImageNet-1K)	72.154
acc@5 (on ImageNet-1K)	90.822
num_params	3504872
min_size	height=1, width=1
categories	tench, goldfish, great white shark, … (997 omitted)
recipe	link

The inference transforms are available at MobileNet_V2_Weights.IMAGENET1K_V2.transforms and perform the following preprocessing operations: Accepts PIL.Image, batched (B, C, H, W) and single (C, H, W) image torch.Tensor objects. The images are resized to resize_size=[232] using interpolation=InterpolationMode.BILINEAR, followed by a central crop of crop_size=[224]. Finally the values are first rescaled to [0.0, 1.0] and then normalized using mean=[0.485, 0.456, 0.406] and std=[0.229, 0.224, 0.225].