Hugging Face's logo

keras
/
dfine_small_coco

KerasHub
Model card Files
xet
Community
prasadsachin commited on
Commit
d1115bb
·
verified ·
1 Parent(s): 98b04b6

Upload folder using huggingface_hub

Browse files
Files changed (7) hide show
  1. README.md +44 -0
  2. config.json +156 -0
  3. image_converter.json +28 -0
  4. metadata.json +9 -0
  5. model.weights.h5 +3 -0
  6. preprocessor.json +46 -0
  7. task.json +246 -0
README.md ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: keras-hub
3
+ ---
4
+ This is a [`DFine` model](https://keras.io/api/keras_hub/models/d_fine) uploaded using the KerasHub library and can be used with JAX, TensorFlow, and PyTorch backends.
5
+ This model is related to a `ObjectDetector` task.
6
+
7
+ Model config:
8
+ * **name:** d_fine_backbone
9
+ * **trainable:** True
10
+ * **backbone:** {'module': 'keras_hub.src.models.hgnetv2.hgnetv2_backbone', 'class_name': 'HGNetV2Backbone', 'config': {'name': 'hg_net_v2_backbone', 'trainable': True, 'depths': [3, 4, 6, 3], 'embedding_size': 32, 'hidden_sizes': [128, 256, 512, 1024], 'stem_channels': [3, 16, 16], 'hidden_act': 'relu', 'use_learnable_affine_block': True, 'stackwise_stage_filters': [[16, 16, 64, 1, 3, 3], [64, 32, 256, 1, 3, 3], [256, 64, 512, 2, 3, 5], [512, 128, 1024, 1, 3, 5]], 'apply_downsample': [False, True, True, True], 'use_lightweight_conv_block': [False, False, True, True], 'image_shape': [None, None, 3], 'out_features': ['stage2', 'stage3', 'stage4'], 'data_format': 'channels_last'}, 'registered_name': 'keras_hub>HGNetV2Backbone'}
11
+ * **decoder_in_channels:** [256, 256, 256]
12
+ * **encoder_hidden_dim:** 256
13
+ * **num_labels:** 80
14
+ * **num_denoising:** 100
15
+ * **learn_initial_query:** False
16
+ * **num_queries:** 300
17
+ * **anchor_image_size:** [640, 640]
18
+ * **feat_strides:** [8, 16, 32]
19
+ * **num_feature_levels:** 3
20
+ * **hidden_dim:** 256
21
+ * **encoder_in_channels:** [256, 512, 1024]
22
+ * **encode_proj_layers:** [2]
23
+ * **num_attention_heads:** 8
24
+ * **encoder_ffn_dim:** 1024
25
+ * **num_encoder_layers:** 1
26
+ * **hidden_expansion:** 0.5
27
+ * **depth_multiplier:** 0.34
28
+ * **eval_idx:** -1
29
+ * **box_noise_scale:** 1.0
30
+ * **label_noise_ratio:** 0.5
31
+ * **labels:** None
32
+ * **num_decoder_layers:** 3
33
+ * **decoder_attention_heads:** 8
34
+ * **decoder_ffn_dim:** 1024
35
+ * **decoder_method:** default
36
+ * **decoder_n_points:** [3, 6, 3]
37
+ * **lqe_hidden_dim:** 64
38
+ * **num_lqe_layers:** 2
39
+ * **seed:** 0
40
+ * **image_shape:** [None, None, 3]
41
+ * **data_format:** channels_last
42
+ * **out_features:** ['stage2', 'stage3', 'stage4']
43
+
44
+ This model card has been generated automatically and should be completed by the model author. See [Model Cards documentation](https://huggingface.co/docs/hub/model-cards) for more information.
config.json ADDED
@@ -0,0 +1,156 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "module": "keras_hub.src.models.d_fine.d_fine_backbone",
3
+ "class_name": "DFineBackbone",
4
+ "config": {
5
+ "name": "d_fine_backbone",
6
+ "trainable": true,
7
+ "backbone": {
8
+ "module": "keras_hub.src.models.hgnetv2.hgnetv2_backbone",
9
+ "class_name": "HGNetV2Backbone",
10
+ "config": {
11
+ "name": "hg_net_v2_backbone",
12
+ "trainable": true,
13
+ "depths": [
14
+ 3,
15
+ 4,
16
+ 6,
17
+ 3
18
+ ],
19
+ "embedding_size": 32,
20
+ "hidden_sizes": [
21
+ 128,
22
+ 256,
23
+ 512,
24
+ 1024
25
+ ],
26
+ "stem_channels": [
27
+ 3,
28
+ 16,
29
+ 16
30
+ ],
31
+ "hidden_act": "relu",
32
+ "use_learnable_affine_block": true,
33
+ "stackwise_stage_filters": [
34
+ [
35
+ 16,
36
+ 16,
37
+ 64,
38
+ 1,
39
+ 3,
40
+ 3
41
+ ],
42
+ [
43
+ 64,
44
+ 32,
45
+ 256,
46
+ 1,
47
+ 3,
48
+ 3
49
+ ],
50
+ [
51
+ 256,
52
+ 64,
53
+ 512,
54
+ 2,
55
+ 3,
56
+ 5
57
+ ],
58
+ [
59
+ 512,
60
+ 128,
61
+ 1024,
62
+ 1,
63
+ 3,
64
+ 5
65
+ ]
66
+ ],
67
+ "apply_downsample": [
68
+ false,
69
+ true,
70
+ true,
71
+ true
72
+ ],
73
+ "use_lightweight_conv_block": [
74
+ false,
75
+ false,
76
+ true,
77
+ true
78
+ ],
79
+ "image_shape": [
80
+ null,
81
+ null,
82
+ 3
83
+ ],
84
+ "out_features": [
85
+ "stage2",
86
+ "stage3",
87
+ "stage4"
88
+ ],
89
+ "data_format": "channels_last"
90
+ },
91
+ "registered_name": "keras_hub>HGNetV2Backbone"
92
+ },
93
+ "decoder_in_channels": [
94
+ 256,
95
+ 256,
96
+ 256
97
+ ],
98
+ "encoder_hidden_dim": 256,
99
+ "num_labels": 80,
100
+ "num_denoising": 100,
101
+ "learn_initial_query": false,
102
+ "num_queries": 300,
103
+ "anchor_image_size": [
104
+ 640,
105
+ 640
106
+ ],
107
+ "feat_strides": [
108
+ 8,
109
+ 16,
110
+ 32
111
+ ],
112
+ "num_feature_levels": 3,
113
+ "hidden_dim": 256,
114
+ "encoder_in_channels": [
115
+ 256,
116
+ 512,
117
+ 1024
118
+ ],
119
+ "encode_proj_layers": [
120
+ 2
121
+ ],
122
+ "num_attention_heads": 8,
123
+ "encoder_ffn_dim": 1024,
124
+ "num_encoder_layers": 1,
125
+ "hidden_expansion": 0.5,
126
+ "depth_multiplier": 0.34,
127
+ "eval_idx": -1,
128
+ "box_noise_scale": 1.0,
129
+ "label_noise_ratio": 0.5,
130
+ "labels": null,
131
+ "num_decoder_layers": 3,
132
+ "decoder_attention_heads": 8,
133
+ "decoder_ffn_dim": 1024,
134
+ "decoder_method": "default",
135
+ "decoder_n_points": [
136
+ 3,
137
+ 6,
138
+ 3
139
+ ],
140
+ "lqe_hidden_dim": 64,
141
+ "num_lqe_layers": 2,
142
+ "seed": 0,
143
+ "image_shape": [
144
+ null,
145
+ null,
146
+ 3
147
+ ],
148
+ "data_format": "channels_last",
149
+ "out_features": [
150
+ "stage2",
151
+ "stage3",
152
+ "stage4"
153
+ ]
154
+ },
155
+ "registered_name": "keras_hub>DFineBackbone"
156
+ }
image_converter.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "module": "keras_hub.src.models.d_fine.d_fine_image_converter",
3
+ "class_name": "DFineImageConverter",
4
+ "config": {
5
+ "name": "d_fine_image_converter",
6
+ "trainable": true,
7
+ "dtype": {
8
+ "module": "keras",
9
+ "class_name": "DTypePolicy",
10
+ "config": {
11
+ "name": "float32"
12
+ },
13
+ "registered_name": null
14
+ },
15
+ "image_size": [
16
+ 640,
17
+ 640
18
+ ],
19
+ "scale": 0.00392156862745098,
20
+ "offset": null,
21
+ "interpolation": "bilinear",
22
+ "antialias": false,
23
+ "crop_to_aspect_ratio": true,
24
+ "pad_to_aspect_ratio": false,
25
+ "bounding_box_format": "yxyx"
26
+ },
27
+ "registered_name": "keras_hub>DFineImageConverter"
28
+ }
metadata.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "keras_version": "3.10.0",
3
+ "keras_hub_version": "0.23.0.dev0",
4
+ "parameter_count": 10329321,
5
+ "date_saved": "2025-08-26@07:48:10",
6
+ "tasks": [
7
+ "ObjectDetector"
8
+ ]
9
+ }
model.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1d1d31dcc49223401a912375cdcbbcee4b19ff594358a518c979d6abd886954d
3
+ size 43140280
preprocessor.json ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "module": "keras_hub.src.models.d_fine.d_fine_object_detector_preprocessor",
3
+ "class_name": "DFineObjectDetectorPreprocessor",
4
+ "config": {
5
+ "name": "d_fine_object_detector_preprocessor",
6
+ "trainable": true,
7
+ "dtype": {
8
+ "module": "keras",
9
+ "class_name": "DTypePolicy",
10
+ "config": {
11
+ "name": "float32"
12
+ },
13
+ "registered_name": null
14
+ },
15
+ "image_converter": {
16
+ "module": "keras_hub.src.models.d_fine.d_fine_image_converter",
17
+ "class_name": "DFineImageConverter",
18
+ "config": {
19
+ "name": "d_fine_image_converter",
20
+ "trainable": true,
21
+ "dtype": {
22
+ "module": "keras",
23
+ "class_name": "DTypePolicy",
24
+ "config": {
25
+ "name": "float32"
26
+ },
27
+ "registered_name": null
28
+ },
29
+ "image_size": [
30
+ 640,
31
+ 640
32
+ ],
33
+ "scale": 0.00392156862745098,
34
+ "offset": null,
35
+ "interpolation": "bilinear",
36
+ "antialias": false,
37
+ "crop_to_aspect_ratio": true,
38
+ "pad_to_aspect_ratio": false,
39
+ "bounding_box_format": "yxyx"
40
+ },
41
+ "registered_name": "keras_hub>DFineImageConverter"
42
+ },
43
+ "config_file": "preprocessor.json"
44
+ },
45
+ "registered_name": "keras_hub>DFineObjectDetectorPreprocessor"
46
+ }
task.json ADDED
@@ -0,0 +1,246 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "module": "keras_hub.src.models.d_fine.d_fine_object_detector",
3
+ "class_name": "DFineObjectDetector",
4
+ "config": {
5
+ "backbone": {
6
+ "module": "keras_hub.src.models.d_fine.d_fine_backbone",
7
+ "class_name": "DFineBackbone",
8
+ "config": {
9
+ "name": "d_fine_backbone",
10
+ "trainable": true,
11
+ "backbone": {
12
+ "module": "keras_hub.src.models.hgnetv2.hgnetv2_backbone",
13
+ "class_name": "HGNetV2Backbone",
14
+ "config": {
15
+ "name": "hg_net_v2_backbone",
16
+ "trainable": true,
17
+ "depths": [
18
+ 3,
19
+ 4,
20
+ 6,
21
+ 3
22
+ ],
23
+ "embedding_size": 32,
24
+ "hidden_sizes": [
25
+ 128,
26
+ 256,
27
+ 512,
28
+ 1024
29
+ ],
30
+ "stem_channels": [
31
+ 3,
32
+ 16,
33
+ 16
34
+ ],
35
+ "hidden_act": "relu",
36
+ "use_learnable_affine_block": true,
37
+ "stackwise_stage_filters": [
38
+ [
39
+ 16,
40
+ 16,
41
+ 64,
42
+ 1,
43
+ 3,
44
+ 3
45
+ ],
46
+ [
47
+ 64,
48
+ 32,
49
+ 256,
50
+ 1,
51
+ 3,
52
+ 3
53
+ ],
54
+ [
55
+ 256,
56
+ 64,
57
+ 512,
58
+ 2,
59
+ 3,
60
+ 5
61
+ ],
62
+ [
63
+ 512,
64
+ 128,
65
+ 1024,
66
+ 1,
67
+ 3,
68
+ 5
69
+ ]
70
+ ],
71
+ "apply_downsample": [
72
+ false,
73
+ true,
74
+ true,
75
+ true
76
+ ],
77
+ "use_lightweight_conv_block": [
78
+ false,
79
+ false,
80
+ true,
81
+ true
82
+ ],
83
+ "image_shape": [
84
+ null,
85
+ null,
86
+ 3
87
+ ],
88
+ "out_features": [
89
+ "stage2",
90
+ "stage3",
91
+ "stage4"
92
+ ],
93
+ "data_format": "channels_last"
94
+ },
95
+ "registered_name": "keras_hub>HGNetV2Backbone"
96
+ },
97
+ "decoder_in_channels": [
98
+ 256,
99
+ 256,
100
+ 256
101
+ ],
102
+ "encoder_hidden_dim": 256,
103
+ "num_labels": 80,
104
+ "num_denoising": 100,
105
+ "learn_initial_query": false,
106
+ "num_queries": 300,
107
+ "anchor_image_size": [
108
+ 640,
109
+ 640
110
+ ],
111
+ "feat_strides": [
112
+ 8,
113
+ 16,
114
+ 32
115
+ ],
116
+ "num_feature_levels": 3,
117
+ "hidden_dim": 256,
118
+ "encoder_in_channels": [
119
+ 256,
120
+ 512,
121
+ 1024
122
+ ],
123
+ "encode_proj_layers": [
124
+ 2
125
+ ],
126
+ "num_attention_heads": 8,
127
+ "encoder_ffn_dim": 1024,
128
+ "num_encoder_layers": 1,
129
+ "hidden_expansion": 0.5,
130
+ "depth_multiplier": 0.34,
131
+ "eval_idx": -1,
132
+ "box_noise_scale": 1.0,
133
+ "label_noise_ratio": 0.5,
134
+ "labels": null,
135
+ "num_decoder_layers": 3,
136
+ "decoder_attention_heads": 8,
137
+ "decoder_ffn_dim": 1024,
138
+ "decoder_method": "default",
139
+ "decoder_n_points": [
140
+ 3,
141
+ 6,
142
+ 3
143
+ ],
144
+ "lqe_hidden_dim": 64,
145
+ "num_lqe_layers": 2,
146
+ "seed": 0,
147
+ "image_shape": [
148
+ null,
149
+ null,
150
+ 3
151
+ ],
152
+ "data_format": "channels_last",
153
+ "out_features": [
154
+ "stage2",
155
+ "stage3",
156
+ "stage4"
157
+ ]
158
+ },
159
+ "registered_name": "keras_hub>DFineBackbone"
160
+ },
161
+ "preprocessor": {
162
+ "module": "keras_hub.src.models.d_fine.d_fine_object_detector_preprocessor",
163
+ "class_name": "DFineObjectDetectorPreprocessor",
164
+ "config": {
165
+ "name": "d_fine_object_detector_preprocessor",
166
+ "trainable": true,
167
+ "dtype": {
168
+ "module": "keras",
169
+ "class_name": "DTypePolicy",
170
+ "config": {
171
+ "name": "float32"
172
+ },
173
+ "registered_name": null
174
+ },
175
+ "image_converter": {
176
+ "module": "keras_hub.src.models.d_fine.d_fine_image_converter",
177
+ "class_name": "DFineImageConverter",
178
+ "config": {
179
+ "name": "d_fine_image_converter",
180
+ "trainable": true,
181
+ "dtype": {
182
+ "module": "keras",
183
+ "class_name": "DTypePolicy",
184
+ "config": {
185
+ "name": "float32"
186
+ },
187
+ "registered_name": null
188
+ },
189
+ "image_size": [
190
+ 640,
191
+ 640
192
+ ],
193
+ "scale": 0.00392156862745098,
194
+ "offset": null,
195
+ "interpolation": "bilinear",
196
+ "antialias": false,
197
+ "crop_to_aspect_ratio": true,
198
+ "pad_to_aspect_ratio": false,
199
+ "bounding_box_format": "yxyx"
200
+ },
201
+ "registered_name": "keras_hub>DFineImageConverter"
202
+ },
203
+ "config_file": "preprocessor.json"
204
+ },
205
+ "registered_name": "keras_hub>DFineObjectDetectorPreprocessor"
206
+ },
207
+ "name": "d_fine_object_detector",
208
+ "num_classes": 80,
209
+ "bounding_box_format": "yxyx",
210
+ "matcher_class_cost": 2.0,
211
+ "matcher_bbox_cost": 5.0,
212
+ "matcher_ciou_cost": 2.0,
213
+ "use_focal_loss": true,
214
+ "matcher_alpha": 0.25,
215
+ "matcher_gamma": 2.0,
216
+ "weight_loss_vfl": 1.0,
217
+ "weight_loss_bbox": 5.0,
218
+ "weight_loss_ciou": 2.0,
219
+ "weight_loss_fgl": 0.15,
220
+ "weight_loss_ddf": 1.5,
221
+ "ddf_temperature": 5.0,
222
+ "prediction_decoder": {
223
+ "module": "keras_hub.src.layers.modeling.non_max_supression",
224
+ "class_name": "NonMaxSuppression",
225
+ "config": {
226
+ "name": "non_max_suppression",
227
+ "trainable": true,
228
+ "dtype": {
229
+ "module": "keras",
230
+ "class_name": "DTypePolicy",
231
+ "config": {
232
+ "name": "float32"
233
+ },
234
+ "registered_name": null
235
+ },
236
+ "bounding_box_format": "yxyx",
237
+ "from_logits": true,
238
+ "iou_threshold": 0.5,
239
+ "confidence_threshold": 0.5,
240
+ "max_detections": 300
241
+ },
242
+ "registered_name": "keras_hub>NonMaxSuppression"
243
+ }
244
+ },
245
+ "registered_name": "keras_hub>DFineObjectDetector"
246
+ }