yologo

package module
v0.0.0-...-cf89635 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Dec 21, 2020 License: Apache-2.0 Imports: 20 Imported by: 0

README

WIP. PRs are welcome

Port of Darknet YOLO, but via Gorgonia Both YOLOv3 and tiny-YOLOv3 are implemented.

Usage

Navigate to example/yolo-v3 folder and run main.go.

Available flags

go run main.go -h

  -cfg string
        Path to net configuration file (default "../../test_network_data/yolov3-tiny.cfg")
  -image string
        Path to image file for 'detector' mode (default "../../test_network_data/dog_416x416.jpg")
  -mode string
        Choose the mode: detector/training (default "detector")
  -train string
        Path to folder with labeled data (default "../../test_yolo_op_data")
  -weights string
        Path to weights file (default "../../test_network_data/yolov3-tiny.weights")

For testing tiny-yolov3:

go run main.go --mode detector --cfg ../../test_network_data/yolov3-tiny.cfg --weights ../../test_network_data/yolov3-tiny.weights --image ../../test_network_data/dog_416x416.jpg

For testing yolov3:

go run main.go --mode detector --cfg ../../test_network_data/yolov3.cfg --weights ../../test_network_data/yolov3.weights --image ../../test_network_data/dog_416x416.jpg

For training WIP. PRs are welcome:

go run main.go --mode training --cfg ../../test_network_data/yolov3-tiny.cfg --weights ../../test_network_data/yolov3-tiny.weights --image ../../test_network_data/dog_416x416.jpg --train ../../test_yolo_op_data

Weights and configuration

Weights can be downloaded via curl-scripts download_weights_yolo_v3.sh and download_weights_yolo_tiny_v3.sh. Configuration files: yolov3-tiny.cfg and yolov3.cfg

Network Architecture

Tiny-YOLOv3 Architecture is:

Convolution layer: Filters->16 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Maxpooling layer: Size->2 Stride->2
Convolution layer: Filters->32 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Maxpooling layer: Size->2 Stride->2
Convolution layer: Filters->64 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Maxpooling layer: Size->2 Stride->2
Convolution layer: Filters->128 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Maxpooling layer: Size->2 Stride->2
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Maxpooling layer: Size->2 Stride->2
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Maxpooling layer: Size->2 Stride->1
Convolution layer: Filters->1024 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->255 Padding->0 Kernel->1x1 Stride->1 Activation->linear Batch->0 Bias->true
YOLO layer: Mask->3 Anchors->[81, 82]   |       Mask->4 Anchors->[135, 169]     |       Mask->5 Anchors->[344, 319]
Route layer: Start->13
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Upsample layer: Scale->2
Route layer: Start->19 End->8
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->255 Padding->0 Kernel->1x1 Stride->1 Activation->linear Batch->0 Bias->true
YOLO layer: Mask->0 Anchors->[10, 14]   |       Mask->1 Anchors->[23, 27]       |       Mask->2 Anchors->[37, 58]

YOLOv3 Architecture is:

Convolution layer: Filters->32 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->64 Padding->1 Kernel->3x3 Stride->2 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->32 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->64 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->1 | index to->3
Convolution layer: Filters->128 Padding->1 Kernel->3x3 Stride->2 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->64 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->128 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->5 | index to->7
Convolution layer: Filters->64 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->128 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->8 | index to->10
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->2 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->12 | index to->14
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->15 | index to->17
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->18 | index to->20
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->21 | index to->23
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->24 | index to->26
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->27 | index to->29
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->30 | index to->32
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->33 | index to->35
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->2 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->37 | index to->39
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->40 | index to->42
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->43 | index to->45
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->46 | index to->48
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->49 | index to->51
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->52 | index to->54
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->55 | index to->57
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->58 | index to->60
Convolution layer: Filters->1024 Padding->1 Kernel->3x3 Stride->2 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->1024 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->62 | index to->64
Convolution layer: Filters->512 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->1024 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->65 | index to->67
Convolution layer: Filters->512 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->1024 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->68 | index to->70
Convolution layer: Filters->512 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->1024 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Shortcut layer: index from->71 | index to->73
Convolution layer: Filters->512 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->1024 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->1024 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->1024 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->255 Padding->0 Kernel->1x1 Stride->1 Activation->linear Batch->0 Bias->true
YOLO layer: Mask->6 Anchors->[116, 90]  |       Mask->7 Anchors->[156, 198]     |       Mask->8 Anchors->[373, 326]
Route layer: Start->79
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Upsample layer: Scale->2
Route layer: Start->85 End->61
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->512 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->255 Padding->0 Kernel->1x1 Stride->1 Activation->linear Batch->0 Bias->true
YOLO layer: Mask->3 Anchors->[30, 61]   |       Mask->4 Anchors->[62, 45]       |       Mask->5 Anchors->[59, 119]
Route layer: Start->91
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Upsample layer: Scale->2
Route layer: Start->97 End->36
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->128 Padding->0 Kernel->1x1 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->256 Padding->1 Kernel->3x3 Stride->1 Activation->leaky Batch->1 Bias->false
Convolution layer: Filters->255 Padding->0 Kernel->1x1 Stride->1 Activation->linear Batch->0 Bias->true
YOLO layer: Mask->0 Anchors->[10, 13]   |       Mask->1 Anchors->[16, 30]       |       Mask->2 Anchors->[33, 23]

Documentation

Index

Constants

This section is empty.

Variables

View Source 
var Slice = gorgonia.S

Slice Just alias

Functions

func Float32frombytes 

func Float32frombytes(bytes []byte) float32

Float32frombytes Converts []byte to float32

func GetFloat32Image 

func GetFloat32Image(fname string, resizeWidth, resizeHeight int) ([]float32, error)

GetFloat32Image Returns []float32 representation of image file

func IOUFloat32 

func IOUFloat32(r1, r2 image.Rectangle) float32

IOUFloat32 Intersection Over Union for float32

func Image2Float32 

func Image2Float32(img image.Image) ([]float32, error)

Image2Float32 Returns []float32 representation of image.Image

func MaxInt 

func MaxInt(a, b int) int

MaxInt Maximum between two integers

func MinInt 

func MinInt(a, b int) int

MinInt Minimum between two integers

func ParseConfiguration 

func ParseConfiguration(fname string) ([]map[string]string, error)

ParseConfiguration Parse darknet configuration file

func ParseWeights 

func ParseWeights(fname string) ([]float32, error)

ParseWeights Parse darknet weights

func Rectify 

func Rectify(x, y, h, w, maxwidth, maxheight int) image.Rectangle

Rectify Creates rectangle

func Upsample2D 

func Upsample2D(x *gorgonia.Node, scale int) (*gorgonia.Node, error)

Upsample2D - simply upscaling Tensor by scale factor. 

1, 2
3, 4
converts to
1,1,2,2
1,1,2,2
3,3,4,4,
3,3,4,4,

Types

type DetectionRectangle 

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

DetectionRectangle Representation of detection

func (*DetectionRectangle) GetClass 

func (dr *DetectionRectangle) GetClass() string

GetClass Returns class of object

func (*DetectionRectangle) String 

func (dr *DetectionRectangle) String() string

type Detections 

type Detections []*DetectionRectangle

Detections Detection rectangles

func (Detections) Len 

func (detections Detections) Len() int

Methods to match sort.Interface interface

func (Detections) Less 

func (detections Detections) Less(i, j int) bool

func (Detections) Swap 

func (detections Detections) Swap(i, j int)

type DetectionsOrder 

type DetectionsOrder []*DetectionRectangle

DetectionsOrder Ordering for X-axis

func (DetectionsOrder) Len 

func (detections DetectionsOrder) Len() int

Methods to match sort.Interface interface

func (DetectionsOrder) Less 

func (detections DetectionsOrder) Less(i, j int) bool

func (DetectionsOrder) Swap 

func (detections DetectionsOrder) Swap(i, j int)

type YOLOv3 

type YOLOv3 struct {
	LearningNodes []*gorgonia.Node
	// contains filtered or unexported fields
}

YOLOv3 YOLOv3 architecture

func NewYoloV3 

func NewYoloV3(g *gorgonia.ExprGraph, input *gorgonia.Node, classesNumber, boxesPerCell int, leakyCoef float64, cfgFile, weightsFile string) (*YOLOv3, error)

NewYoloV3 Create new YOLO v3

func (*YOLOv3) ActivateTrainingMode 

func (net *YOLOv3) ActivateTrainingMode() error

ActivateTrainingMode Activates training mode for unexported yoloOP

func (*YOLOv3) DisableTrainingMode 

func (net *YOLOv3) DisableTrainingMode() error

DisableTrainingMode Disables training mode for unexported yoloOP

func (*YOLOv3) GetOutput 

func (net *YOLOv3) GetOutput() []*gorgonia.Node

GetOutput Get out YOLO layers (can be multiple of them)

func (*YOLOv3) Print 

func (net *YOLOv3) Print()

Print Print architecture of network

func (*YOLOv3) ProcessOutput 

func (net *YOLOv3) ProcessOutput(classes []string, scoreTreshold, iouTreshold float32) (Detections, error)

ProcessOutput Returns postprocessed detections

func (*YOLOv3) SetTarget 

func (net *YOLOv3) SetTarget(target []float32) error

SetTarget Set desired target for net's output (for training mode)

type YoloTrainer 

type YoloTrainer interface {
	ActivateTrainingMode()
	DisableTrainingMode()
	SetTarget([]float32)
}

YoloTrainer Wrapper around yoloOP It has method for setting desired bboxes as output of network

func YOLOv3Node 

func YOLOv3Node(input *gorgonia.Node, anchors []float32, masks []int, netSize, numClasses int, ignoreTresh float32, targets ...*gorgonia.Node) (*gorgonia.Node, YoloTrainer, error)

YOLOv3Node Constructor for YOLO-based node operation 

input - Input node
anchors - Slice of anchors
masks - Slice of masks
netSize - Height/Width of input
numClasses - Amount of classes
ignoreTresh - Treshold
targets - Desired targets.

Source Files

View all Source files

Directories

Path Synopsis
example
yolo-v3

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.