-
Notifications
You must be signed in to change notification settings - Fork 110
/
vision.go
209 lines (192 loc) · 7.08 KB
/
vision.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
// Package vision is the service that allows you to access various computer vision algorithms
// (like detection, segmentation, tracking, etc) that usually only require a camera or image input.
package vision
import (
"context"
"image"
"github.com/pkg/errors"
"go.opencensus.io/trace"
servicepb "go.viam.com/api/service/vision/v1"
"go.viam.com/rdk/components/camera"
"go.viam.com/rdk/resource"
"go.viam.com/rdk/robot"
viz "go.viam.com/rdk/vision"
"go.viam.com/rdk/vision/classification"
"go.viam.com/rdk/vision/objectdetection"
"go.viam.com/rdk/vision/segmentation"
)
func init() {
resource.RegisterAPI(API, resource.APIRegistration[Service]{
RPCServiceServerConstructor: NewRPCServiceServer,
RPCServiceHandler: servicepb.RegisterVisionServiceHandlerFromEndpoint,
RPCServiceDesc: &servicepb.VisionService_ServiceDesc,
RPCClient: NewClientFromConn,
})
}
// A Service that implements various computer vision algorithms like detection and segmentation.
type Service interface {
resource.Resource
DetectionsFromCamera(ctx context.Context, cameraName string, extra map[string]interface{}) ([]objectdetection.Detection, error)
Detections(ctx context.Context, img image.Image, extra map[string]interface{}) ([]objectdetection.Detection, error)
// classifier methods
ClassificationsFromCamera(
ctx context.Context,
cameraName string,
n int,
extra map[string]interface{},
) (classification.Classifications, error)
Classifications(
ctx context.Context,
img image.Image,
n int,
extra map[string]interface{},
) (classification.Classifications, error)
// segmenter methods
GetObjectPointClouds(ctx context.Context, cameraName string, extra map[string]interface{}) ([]*viz.Object, error)
}
// SubtypeName is the name of the type of service.
const SubtypeName = "vision"
// API is a variable that identifies the vision service resource API.
var API = resource.APINamespaceRDK.WithServiceType(SubtypeName)
// Named is a helper for getting the named vision's typed resource name.
// RSDK-347 Implements vision's Named.
func Named(name string) resource.Name {
return resource.NewName(API, name)
}
// FromRobot is a helper for getting the named vision service from the given Robot.
func FromRobot(r robot.Robot, name string) (Service, error) {
return robot.ResourceFromRobot[Service](r, Named(name))
}
// vizModel wraps the vision model with all the service interface methods.
type vizModel struct {
resource.Named
resource.AlwaysRebuild
resource.TriviallyCloseable
name string
r robot.Robot // in order to get access to all cameras
closerFunc func(ctx context.Context) error // close the underlying model
classifierFunc classification.Classifier
detectorFunc objectdetection.Detector
segmenter3DFunc segmentation.Segmenter
}
// NewService wraps the vision model in the struct that fulfills the vision service interface.
func NewService(
name string,
r robot.Robot,
c func(ctx context.Context) error,
cf classification.Classifier,
df objectdetection.Detector,
s3f segmentation.Segmenter,
) (Service, error) {
if cf == nil && df == nil && s3f == nil {
return nil, errors.Errorf(
"model %q does not fulfill any method of the vision service. It is neither a detector, nor classifier, nor 3D segmenter", name)
}
return &vizModel{
name: name,
r: r,
closerFunc: c,
classifierFunc: cf,
detectorFunc: df,
segmenter3DFunc: s3f,
}, nil
}
// Detections returns the detections of given image if the model implements objectdetector.Detector.
func (vm *vizModel) Detections(
ctx context.Context,
img image.Image,
extra map[string]interface{},
) ([]objectdetection.Detection, error) {
ctx, span := trace.StartSpan(ctx, "service::vision::Detections::"+vm.name)
defer span.End()
if vm.detectorFunc == nil {
return nil, errors.Errorf("vision model %q does not implement a Detector", vm.name)
}
return vm.detectorFunc(ctx, img)
}
// DetectionsFromCamera returns the detections of the next image from the given camera.
func (vm *vizModel) DetectionsFromCamera(
ctx context.Context,
cameraName string,
extra map[string]interface{},
) ([]objectdetection.Detection, error) {
ctx, span := trace.StartSpan(ctx, "service::vision::DetectionsFromCamera::"+vm.name)
defer span.End()
if vm.detectorFunc == nil {
return nil, errors.Errorf("vision model %q does not implement a Detector", vm.name)
}
cam, err := camera.FromRobot(vm.r, cameraName)
if err != nil {
return nil, errors.Wrapf(err, "could not find camera named %s", cameraName)
}
img, release, err := camera.ReadImage(ctx, cam)
if err != nil {
return nil, errors.Wrapf(err, "could not get image from %s", cameraName)
}
defer release()
return vm.detectorFunc(ctx, img)
}
// Classifications returns the classifications of given image if the model implements classifications.Classifier.
func (vm *vizModel) Classifications(
ctx context.Context,
img image.Image,
n int,
extra map[string]interface{},
) (classification.Classifications, error) {
ctx, span := trace.StartSpan(ctx, "service::vision::Classifications::"+vm.name)
defer span.End()
if vm.classifierFunc == nil {
return nil, errors.Errorf("vision model %q does not implement a Classifier", vm.name)
}
fullClassifications, err := vm.classifierFunc(ctx, img)
if err != nil {
return nil, errors.Wrap(err, "could not get classifications from image")
}
return fullClassifications.TopN(n)
}
// ClassificationsFromCamera returns the classifications of the next image from the given camera.
func (vm *vizModel) ClassificationsFromCamera(
ctx context.Context,
cameraName string,
n int,
extra map[string]interface{},
) (classification.Classifications, error) {
ctx, span := trace.StartSpan(ctx, "service::vision::ClassificationsFromCamera::"+vm.name)
defer span.End()
if vm.classifierFunc == nil {
return nil, errors.Errorf("vision model %q does not implement a Classifier", vm.name)
}
cam, err := camera.FromRobot(vm.r, cameraName)
if err != nil {
return nil, errors.Wrapf(err, "could not find camera named %s", cameraName)
}
img, release, err := camera.ReadImage(ctx, cam)
if err != nil {
return nil, errors.Wrapf(err, "could not get image from %s", cameraName)
}
defer release()
fullClassifications, err := vm.classifierFunc(ctx, img)
if err != nil {
return nil, errors.Wrap(err, "could not get classifications from image")
}
return fullClassifications.TopN(n)
}
// GetObjectPointClouds returns all the found objects in a 3D image if the model implements Segmenter3D.
func (vm *vizModel) GetObjectPointClouds(ctx context.Context, cameraName string, extra map[string]interface{}) ([]*viz.Object, error) {
if vm.segmenter3DFunc == nil {
return nil, errors.Errorf("vision model %q does not implement a 3D segmenter", vm.name)
}
ctx, span := trace.StartSpan(ctx, "service::vision::GetObjectPointClouds::"+vm.name)
defer span.End()
cam, err := camera.FromRobot(vm.r, cameraName)
if err != nil {
return nil, err
}
return vm.segmenter3DFunc(ctx, cam)
}
func (vm *vizModel) Close(ctx context.Context) error {
if vm.closerFunc == nil {
return nil
}
return vm.closerFunc(ctx)
}