vision

package

v0.9.1 Latest Latest Go to latest Published: Apr 21, 2024 License: Apache-2.0 Imports: 13 Imported by: 0

Details

Valid go.mod file
Redistributable license
Tagged version
Stable version
Learn more about best practices

Repository

git.andr3h3nriqu3s.com/andr3/gotch

Documentation ¶

Index ¶

func AlexNet(p *nn.Path, nclasses int64) ts.ModuleT
func Augmentation(t *ts.Tensor, flip bool, crop int64, cutout int64) *ts.Tensor
func DenseNet121(p *nn.Path, nclasses int64) ts.ModuleT
func DenseNet161(p *nn.Path, nclasses int64) ts.ModuleT
func DenseNet169(p *nn.Path, nclasses int64) ts.ModuleT
func DenseNet201(p *nn.Path, nclasses int64) ts.ModuleT
func EfficientNetB0(p *nn.Path, nclasses int64) ts.ModuleT
func EfficientNetB1(p *nn.Path, nclasses int64) ts.ModuleT
func EfficientNetB2(p *nn.Path, nclasses int64) ts.ModuleT
func EfficientNetB3(p *nn.Path, nclasses int64) ts.ModuleT
func EfficientNetB4(p *nn.Path, nclasses int64) ts.ModuleT
func EfficientNetB5(p *nn.Path, nclasses int64) ts.ModuleT
func EfficientNetB6(p *nn.Path, nclasses int64) ts.ModuleT
func EfficientNetB7(p *nn.Path, nclasses int64) ts.ModuleT
func InceptionV3(p *nn.Path, nclasses int64) ts.ModuleT
func Load(path string) (*ts.Tensor, error)
func LoadAndResize(path string, outW int64, outH int64) (*ts.Tensor, error)
func LoadDir(dir string, outW int64, outH int64) (*ts.Tensor, error)
func MobileNetV2(p *nn.Path, nclasses int64) ts.ModuleT
func RandomCrop(t *ts.Tensor, pad int64) *ts.Tensor
func RandomCutout(t *ts.Tensor, sz int64) *ts.Tensor
func RandomFlip(t *ts.Tensor) *ts.Tensor
func ResNet101(path *nn.Path, numClasses int64) ts.ModuleT
func ResNet101NoFinalLayer(path *nn.Path) ts.ModuleT
func ResNet150NoFinalLayer(path *nn.Path) ts.ModuleT
func ResNet152(path *nn.Path, numClasses int64) ts.ModuleT
func ResNet18(path *nn.Path, numClasses int64) nn.FuncT
func ResNet18NoFinalLayer(path *nn.Path) nn.FuncT
func ResNet34(path *nn.Path, numClasses int64) nn.FuncT
func ResNet34NoFinalLayer(path *nn.Path) nn.FuncT
func ResNet50(path *nn.Path, numClasses int64) ts.ModuleT
func ResNet50NoFinalLayer(path *nn.Path) ts.ModuleT
func Resize(t *ts.Tensor, outW int64, outH int64) (*ts.Tensor, error)
func ResizePreserveAspectRatio(t *ts.Tensor, outW int64, outH int64) (*ts.Tensor, error)
func Save(tensor *ts.Tensor, path string) error
func SqueezeNetV1_0(p *nn.Path, nclasses int64) ts.ModuleT
func SqueezeNetV1_1(p *nn.Path, nclasses int64) ts.ModuleT
func VGG11(path *nn.Path, nclasses int64) *nn.SequentialT
func VGG11BN(path *nn.Path, nclasses int64) *nn.SequentialT
func VGG13(path *nn.Path, nclasses int64) *nn.SequentialT
func VGG13BN(path *nn.Path, nclasses int64) *nn.SequentialT
func VGG16(path *nn.Path, nclasses int64) *nn.SequentialT
func VGG16BN(path *nn.Path, nclasses int64) *nn.SequentialT
func VGG19(path *nn.Path, nclasses int64) *nn.SequentialT
func VGG19BN(path *nn.Path, nclasses int64) *nn.SequentialT
type BlockArgs
type Dataset
- func CFLoadDir(dir string) *Dataset
- func LoadMNISTDir(dir string) *Dataset
- func (ds *Dataset) TestIter(batchSize int64) *ts.Iter2
- func (ds *Dataset) TrainIter(batchSize int64) *ts.Iter2
type ImageNet
- func NewImageNet() *ImageNet
- func (in *ImageNet) ClassCount() int64
- func (in *ImageNet) Classes() []string
- func (in *ImageNet) LoadFromDir(path string) (*Dataset, error)
- func (in *ImageNet) LoadImage(path string) (*ts.Tensor, error)
- func (in *ImageNet) LoadImageAndResize(path string, w, h int64) (*ts.Tensor, error)
- func (in *ImageNet) LoadImageAndResize224(path string) (*ts.Tensor, error)
- func (in *ImageNet) Normalize(tensor *ts.Tensor) (*ts.Tensor, error)
- func (in *ImageNet) SaveImage(tensor *ts.Tensor, path string) error
- func (in *ImageNet) Top(input *ts.Tensor, k int64) []TopItem
- func (in *ImageNet) UnNormalize(tensor *ts.Tensor) (*ts.Tensor, error)
type TopItem

Constants ¶

This section is empty.

Variables ¶

This section is empty.

Functions ¶

func AlexNet ¶

func AlexNet(p *nn.Path, nclasses int64) ts.ModuleT

func Augmentation ¶

func Augmentation(t *ts.Tensor, flip bool, crop int64, cutout int64) *ts.Tensor

func DenseNet121 ¶

func DenseNet121(p *nn.Path, nclasses int64) ts.ModuleT

func DenseNet161 ¶

func DenseNet161(p *nn.Path, nclasses int64) ts.ModuleT

func DenseNet169 ¶

func DenseNet169(p *nn.Path, nclasses int64) ts.ModuleT

func DenseNet201 ¶

func DenseNet201(p *nn.Path, nclasses int64) ts.ModuleT

func EfficientNetB0 ¶

func EfficientNetB0(p *nn.Path, nclasses int64) ts.ModuleT

func EfficientNetB1 ¶

func EfficientNetB1(p *nn.Path, nclasses int64) ts.ModuleT

func EfficientNetB2 ¶

func EfficientNetB2(p *nn.Path, nclasses int64) ts.ModuleT

func EfficientNetB3 ¶

func EfficientNetB3(p *nn.Path, nclasses int64) ts.ModuleT

func EfficientNetB4 ¶

func EfficientNetB4(p *nn.Path, nclasses int64) ts.ModuleT

func EfficientNetB5 ¶

func EfficientNetB5(p *nn.Path, nclasses int64) ts.ModuleT

func EfficientNetB6 ¶

func EfficientNetB6(p *nn.Path, nclasses int64) ts.ModuleT

func EfficientNetB7 ¶

func EfficientNetB7(p *nn.Path, nclasses int64) ts.ModuleT

func InceptionV3 ¶

func InceptionV3(p *nn.Path, nclasses int64) ts.ModuleT

func Load ¶

func Load(path string) (*ts.Tensor, error)

Load loads an image from a file.

On success returns a tensor of shape [channel, height, width].

func LoadAndResize ¶

func LoadAndResize(path string, outW int64, outH int64) (*ts.Tensor, error)

LoadAndResize loads and resizes an image, preserve the aspect ratio by taking a center crop.

func LoadDir ¶

func LoadDir(dir string, outW int64, outH int64) (*ts.Tensor, error)

LoadDir loads all the images in a directory.

func MobileNetV2 ¶

func MobileNetV2(p *nn.Path, nclasses int64) ts.ModuleT

func RandomCrop ¶

func RandomCrop(t *ts.Tensor, pad int64) *ts.Tensor

Pad the image using reflections and take some random crops. This expects a 4 dimension NCHW tensor and returns a tensor with an identical shape.

func RandomCutout ¶

func RandomCutout(t *ts.Tensor, sz int64) *ts.Tensor

Applies cutout: randomly remove some square areas in the original images. https://arxiv.org/abs/1708.04552

func RandomFlip ¶

func RandomFlip(t *ts.Tensor) *ts.Tensor

RandomFlip randomly applies horizontal flips This expects a 4 dimension NCHW tensor and returns a tensor with an identical shape.

func ResNet101NoFinalLayer ¶

func ResNet101NoFinalLayer(path *nn.Path) ts.ModuleT

ResNet101 creates a ResNet-101 model without final fully connfected layer.

func ResNet150NoFinalLayer ¶

func ResNet150NoFinalLayer(path *nn.Path) ts.ModuleT

ResNet150 creates a ResNet-150 model without final fully connfected layer.

func ResNet18NoFinalLayer ¶

func ResNet18NoFinalLayer(path *nn.Path) nn.FuncT

ResNet18 creates a ResNet-18 model without final fully connfected layer.

func ResNet34NoFinalLayer ¶

func ResNet34NoFinalLayer(path *nn.Path) nn.FuncT

ResNet34 creates a ResNet-34 model without final fully connfected layer.

func ResNet50NoFinalLayer ¶

func ResNet50NoFinalLayer(path *nn.Path) ts.ModuleT

ResNet50 creates a ResNet-50 model without final fully connfected layer.

func Resize ¶

func Resize(t *ts.Tensor, outW int64, outH int64) (*ts.Tensor, error)

Resize resizes an image.

This expects as input a tensor of shape [channel, height, width] and returns a tensor of shape [channel, out_h, out_w].

func ResizePreserveAspectRatio ¶

func ResizePreserveAspectRatio(t *ts.Tensor, outW int64, outH int64) (*ts.Tensor, error)

ResizePreserveAspectRatio resizes an image, preserve the aspect ratio by taking a center crop.

This expects as input a tensor of shape [channel, height, width] and returns

This expects as input a tensor of shape [channel, height, width]. The image format is based on the filename suffix, supported suffixes are jpg, png, tga, and bmp. The tensor input should be of kind UInt8 with values ranging from 0 to 255.

func SqueezeNetV1_0 ¶

func SqueezeNetV1_0(p *nn.Path, nclasses int64) ts.ModuleT

func SqueezeNetV1_1 ¶

func SqueezeNetV1_1(p *nn.Path, nclasses int64) ts.ModuleT

func VGG11 ¶

func VGG11(path *nn.Path, nclasses int64) *nn.SequentialT

func VGG11BN ¶

func VGG11BN(path *nn.Path, nclasses int64) *nn.SequentialT

func VGG13 ¶

func VGG13(path *nn.Path, nclasses int64) *nn.SequentialT

func VGG13BN ¶

func VGG13BN(path *nn.Path, nclasses int64) *nn.SequentialT

func VGG16 ¶

func VGG16(path *nn.Path, nclasses int64) *nn.SequentialT

func VGG16BN ¶

func VGG16BN(path *nn.Path, nclasses int64) *nn.SequentialT

func VGG19 ¶

func VGG19(path *nn.Path, nclasses int64) *nn.SequentialT

func VGG19BN ¶

func VGG19BN(path *nn.Path, nclasses int64) *nn.SequentialT

Types ¶

type BlockArgs ¶

type BlockArgs struct {
	KernelSize   int64
	NumRepeat    int64
	InputFilters int64
	OutputFilter int64
	ExpandRatio  int64
	SeRatio      float64
	Stride       int64
}

type Dataset ¶

type Dataset struct {
	TrainImages *ts.Tensor
	TrainLabels *ts.Tensor
	TestImages  *ts.Tensor
	TestLabels  *ts.Tensor
	Labels      int64
}

func CFLoadDir ¶

func CFLoadDir(dir string) *Dataset

func LoadMNISTDir ¶

func LoadMNISTDir(dir string) *Dataset

LoadMNISTDir loads all MNIST data from a given directory to Dataset

func (*Dataset) TestIter ¶

func (ds *Dataset) TestIter(batchSize int64) *ts.Iter2

TestIter creates an iterator of Iter type for test images and labels

func (*Dataset) TrainIter ¶

func (ds *Dataset) TrainIter(batchSize int64) *ts.Iter2

TrainIter creates an iterator of Iter type for train images and labels

type ImageNet ¶

type ImageNet struct {
	// contains filtered or unexported fields
}

func NewImageNet ¶

func NewImageNet() *ImageNet

func (*ImageNet) ClassCount ¶

func (in *ImageNet) ClassCount() int64

func (*ImageNet) Classes ¶

func (in *ImageNet) Classes() []string

func (*ImageNet) LoadFromDir ¶

func (in *ImageNet) LoadFromDir(path string) (*Dataset, error)

LoadFromDir loads a dataset from a directory.

NOTE: This assumes that the directory contains two subdirectories named train and val. In each of these datasets, there should be a subdirectory per class named in the same way. The ImageNet normalization is applied, image are resized to 224x224.

func (*ImageNet) LoadImage ¶

func (in *ImageNet) LoadImage(path string) (*ts.Tensor, error)

Load loads an image from a file and applies the ImageNet normalization.

func (*ImageNet) LoadImageAndResize ¶

func (in *ImageNet) LoadImageAndResize(path string, w, h int64) (*ts.Tensor, error)

LoadImageAndResize loads an image from a file and resize it to the specified width and height.

NOTE: This will apply the ImageNet normalization.

func (*ImageNet) LoadImageAndResize224 ¶

func (in *ImageNet) LoadImageAndResize224(path string) (*ts.Tensor, error)

LoadImageAndResize224 loads an image from a file and resize it to 224x224.