-
Notifications
You must be signed in to change notification settings - Fork 110
/
registry.go
237 lines (211 loc) · 7.21 KB
/
registry.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
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
//go:build !arm && !windows
package builtin
import (
"context"
"io"
"github.com/edaniels/golog"
"github.com/invopop/jsonschema"
"github.com/pkg/errors"
"go.opencensus.io/trace"
"go.uber.org/multierr"
"go.viam.com/rdk/services/vision"
"go.viam.com/rdk/vision/classification"
"go.viam.com/rdk/vision/objectdetection"
"go.viam.com/rdk/vision/segmentation"
)
// VisOperation defines what types of operations are allowed by the vision service.
type VisOperation string
// The set of allowed vision model types.
const (
TFLiteDetector = vision.VisModelType("tflite_detector")
TFDetector = vision.VisModelType("tf_detector")
ColorDetector = vision.VisModelType("color_detector")
TFLiteClassifier = vision.VisModelType("tflite_classifier")
TFClassifier = vision.VisModelType("tf_classifier")
RCSegmenter = vision.VisModelType("radius_clustering_segmenter")
DetectorSegmenter = vision.VisModelType("detector_segmenter")
)
// registeredModelParameterSchemas maps the vision model types to the necessary parameters needed to create them.
var registeredModelParameterSchemas = map[vision.VisModelType]*jsonschema.Schema{
TFLiteDetector: jsonschema.Reflect(&TFLiteDetectorConfig{}),
ColorDetector: jsonschema.Reflect(&objectdetection.ColorDetectorConfig{}),
TFLiteClassifier: jsonschema.Reflect(&TFLiteClassifierConfig{}),
RCSegmenter: jsonschema.Reflect(&segmentation.RadiusClusteringConfig{}),
DetectorSegmenter: jsonschema.Reflect(&segmentation.DetectionSegmenterConfig{}),
}
// The set of operations supported by the vision model types.
const (
VisDetection = VisOperation("detection")
VisClassification = VisOperation("classification")
VisSegmentation = VisOperation("segmentation")
)
// visModelToOpMap maps the vision model type with the corresponding vision operation.
var visModelToOpMap = map[vision.VisModelType]VisOperation{
TFLiteDetector: VisDetection,
TFDetector: VisDetection,
ColorDetector: VisDetection,
TFLiteClassifier: VisClassification,
TFClassifier: VisClassification,
RCSegmenter: VisSegmentation,
DetectorSegmenter: VisSegmentation,
}
// newVisModelTypeNotImplemented is used when the model type is not implemented.
func newVisModelTypeNotImplemented(name string) error {
return errors.Errorf("vision model type %q is not implemented", name)
}
type modelMap map[string]registeredModel
// registeredModel struct that holds models parameters.
type registeredModel struct {
Model interface{}
ModelType vision.VisModelType
Closer io.Closer
}
// ToDetector converts model to a dectector.
func (m *registeredModel) toDetector() (objectdetection.Detector, error) {
toReturn, ok := m.Model.(objectdetection.Detector)
if !ok {
return nil, errors.New("couldn't convert model to detector")
}
return toReturn, nil
}
// ToClassifier converts model to a classifier.
func (m *registeredModel) toClassifier() (classification.Classifier, error) {
toReturn, ok := m.Model.(classification.Classifier)
if !ok {
return nil, errors.New("couldn't convert model to classifier")
}
return toReturn, nil
}
// ToSegmenter concerts model to a segmenter.
func (m *registeredModel) toSegmenter() (segmentation.Segmenter, error) {
toReturn, ok := m.Model.(segmentation.Segmenter)
if !ok {
return nil, errors.New("couldn't convert model to segmenter")
}
return toReturn, nil
}
// DetectorNames returns list copy of all detector names.
func (mm modelMap) DetectorNames() []string {
names := make([]string, 0, len(mm))
for name := range mm {
thisType, err := mm.getModelType(name)
if err == nil { // found the model
if visModelToOpMap[thisType] == VisDetection {
names = append(names, name)
}
}
}
return names
}
// ClassifierNames returns a list copy of all classifier names.
func (mm modelMap) ClassifierNames() []string {
names := make([]string, 0, len(mm))
for name := range mm {
thisType, err := mm.getModelType(name)
if err == nil {
if visModelToOpMap[thisType] == VisClassification {
names = append(names, name)
}
}
}
return names
}
// SegmenterNames returns a list copy of all segmenter names.
func (mm modelMap) SegmenterNames() []string {
names := make([]string, 0, len(mm))
for name := range mm {
thisType, err := mm.getModelType(name)
if err == nil {
if visModelToOpMap[thisType] == VisSegmentation {
names = append(names, name)
}
}
}
return names
}
func (mm modelMap) getModelType(name string) (vision.VisModelType, error) {
m, ok := mm[name]
if !ok {
return "", errors.Errorf("no such vision model with name %q", name)
}
return m.ModelType, nil
}
// modelLookup checks to see if model is valid.
func (mm modelMap) modelLookup(name string) (registeredModel, error) {
m, ok := mm[name]
if !ok {
return registeredModel{}, errors.Errorf("no such vision model with name %q", name)
}
return m, nil
}
// ModelNames returns an array copy of all model names.
func (mm modelMap) ModelNames() []string {
names := make([]string, 0, len(mm))
for name := range mm {
names = append(names, name)
}
return names
}
// removeVisModel removes models from valid models.
func (mm modelMap) removeVisModel(name string, logger golog.Logger) error {
if _, ok := mm[name]; !ok {
logger.Infof("no such vision model with name %s", name)
return nil
}
if mm[name].Closer != nil {
err := mm[name].Closer.Close()
if err != nil {
return err
}
}
delete(mm, name)
return nil
}
// RegisterVisModel registers a new model.
func (mm modelMap) RegisterVisModel(name string, m *registeredModel, logger golog.Logger) error {
if m == nil || m.Model == nil {
return errors.Errorf("cannot register a nil model: %s", name)
}
if m.Closer != nil {
mm[name] = registeredModel{
Model: m.Model, ModelType: m.ModelType, Closer: m.Closer,
}
return nil
}
if _, old := mm[name]; old {
logger.Infof("overwriting the model with name: %s", name)
}
mm[name] = registeredModel{
Model: m.Model, ModelType: m.ModelType, Closer: nil,
}
return nil
}
// registerNewVisModels take an attributes struct and parses each element by type to create an RDK Detector
// and register it to the detector map.
func registerNewVisModels(ctx context.Context, mm modelMap, attrs *vision.Attributes, logger golog.Logger) error {
_, span := trace.StartSpan(ctx, "service::vision::registerNewVisModels")
defer span.End()
var err error
for _, attr := range attrs.ModelRegistry {
logger.Debugf("adding vision model %q of type %q", attr.Name, attr.Type)
switch vision.VisModelType(attr.Type) {
case TFLiteDetector:
multierr.AppendInto(&err, registerTfliteDetector(ctx, mm, &attr, logger))
case TFLiteClassifier:
multierr.AppendInto(&err, registerTfliteClassifier(ctx, mm, &attr, logger))
case TFDetector:
multierr.AppendInto(&err, newVisModelTypeNotImplemented(attr.Type))
case TFClassifier:
multierr.AppendInto(&err, newVisModelTypeNotImplemented(attr.Type))
case ColorDetector:
multierr.AppendInto(&err, registerColorDetector(ctx, mm, &attr, logger))
case RCSegmenter:
multierr.AppendInto(&err, registerRCSegmenter(ctx, mm, &attr, logger))
case DetectorSegmenter:
multierr.AppendInto(&err, registerSegmenterFromDetector(ctx, mm, &attr, logger))
default:
multierr.AppendInto(&err, newVisModelTypeNotImplemented(attr.Type))
}
}
return err
}