resnext101_32x8d

torchvision.models.quantization.resnext101_32x8d(*, weights: Optional[Union[ResNeXt101_32X8D_QuantizedWeights, ResNeXt101_32X8D_Weights]] = None, progress: bool = True, quantize: bool = False, **kwargs: Any) → QuantizableResNet

ResNeXt-101 32x8d model from Aggregated Residual Transformation for Deep Neural Networks

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 (ResNeXt101_32X8D_QuantizedWeights or ResNeXt101_32X8D_Weights, optional) – The pretrained weights for the model. See ResNet101_32X8D_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, return a quantized version of the model. Default is False.

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

class torchvision.models.quantization.ResNeXt101_32X8D_QuantizedWeights(value)

The model builder above accepts the following values as the weights parameter. ResNeXt101_32X8D_QuantizedWeights.DEFAULT is equivalent to ResNeXt101_32X8D_QuantizedWeights.IMAGENET1K_FBGEMM_V2. You can also use strings, e.g. weights='DEFAULT' or weights='IMAGENET1K_FBGEMM_V1'.

ResNeXt101_32X8D_QuantizedWeights.IMAGENET1K_FBGEMM_V1:

These weights were produced by doing Post Training Quantization (eager mode) on top of the unquantized weights listed below.

acc@1 (on ImageNet-1K)	78.986
acc@5 (on ImageNet-1K)	94.48
min_size	height=1, width=1
categories	tench, goldfish, great white shark, … (997 omitted)
backend	fbgemm
recipe	link
num_params	88791336
unquantized	ResNeXt101_32X8D_Weights.IMAGENET1K_V1
GIPS	16.41
File size	86.0 MB

The inference transforms are available at ResNeXt101_32X8D_QuantizedWeights.IMAGENET1K_FBGEMM_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].

ResNeXt101_32X8D_QuantizedWeights.IMAGENET1K_FBGEMM_V2:

These weights were produced by doing Post Training Quantization (eager mode) on top of the unquantized weights listed below. Also available as ResNeXt101_32X8D_QuantizedWeights.DEFAULT.

acc@1 (on ImageNet-1K)	82.574
acc@5 (on ImageNet-1K)	96.132
min_size	height=1, width=1
categories	tench, goldfish, great white shark, … (997 omitted)
backend	fbgemm
recipe	link
num_params	88791336
unquantized	ResNeXt101_32X8D_Weights.IMAGENET1K_V2
GIPS	16.41
File size	86.6 MB

The inference transforms are available at ResNeXt101_32X8D_QuantizedWeights.IMAGENET1K_FBGEMM_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].

class torchvision.models.ResNeXt101_32X8D_Weights(value)

The model builder above accepts the following values as the weights parameter. ResNeXt101_32X8D_Weights.DEFAULT is equivalent to ResNeXt101_32X8D_Weights.IMAGENET1K_V2. You can also use strings, e.g. weights='DEFAULT' or weights='IMAGENET1K_V1'.

ResNeXt101_32X8D_Weights.IMAGENET1K_V1:

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

acc@1 (on ImageNet-1K)	79.312
acc@5 (on ImageNet-1K)	94.526
min_size	height=1, width=1
categories	tench, goldfish, great white shark, … (997 omitted)
num_params	88791336
recipe	link
GFLOPS	16.41
File size	339.6 MB

The inference transforms are available at ResNeXt101_32X8D_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].

ResNeXt101_32X8D_Weights.IMAGENET1K_V2:

These weights improve upon the results of the original paper by using TorchVision’s new training recipe. Also available as ResNeXt101_32X8D_Weights.DEFAULT.

acc@1 (on ImageNet-1K)	82.834
acc@5 (on ImageNet-1K)	96.228
min_size	height=1, width=1
categories	tench, goldfish, great white shark, … (997 omitted)
num_params	88791336
recipe	link
GFLOPS	16.41
File size	339.7 MB

The inference transforms are available at ResNeXt101_32X8D_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].