Hugging Face's logo

onnx-community
/
sam3-tracker-ONNX

Mask Generation
Transformers.js
ONNX
sam3_tracker
sam3
Model card Files
xet
 Community
1
Xenova HF Staff commited on
Commit
3460ae6
·
verified ·
1 Parent(s): 861a3ee

Create config.json

Browse files
Files changed (1) hide show
  1. config.json +183 -0
config.json ADDED
@@ -0,0 +1,183 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "enable_occlusion_spatial_embedding": true,
3
+ "enable_temporal_pos_encoding_for_object_pointers": true,
4
+ "image_size": 1008,
5
+ "initializer_range": 0.02,
6
+ "mask_decoder_config": {
7
+ "attention_downsample_rate": 2,
8
+ "dynamic_multimask_stability_delta": 0.05,
9
+ "dynamic_multimask_stability_thresh": 0.98,
10
+ "dynamic_multimask_via_stability": true,
11
+ "hidden_act": "gelu",
12
+ "hidden_size": 256,
13
+ "iou_head_depth": 3,
14
+ "iou_head_hidden_dim": 256,
15
+ "mlp_dim": 2048,
16
+ "model_type": "",
17
+ "num_attention_heads": 8,
18
+ "num_hidden_layers": 2,
19
+ "num_multimask_outputs": 3
20
+ },
21
+ "mask_downsampler_embed_dim": 256,
22
+ "mask_downsampler_hidden_act": "gelu",
23
+ "mask_downsampler_kernel_size": 3,
24
+ "mask_downsampler_padding": 1,
25
+ "mask_downsampler_stride": 2,
26
+ "mask_downsampler_total_stride": 16,
27
+ "max_cond_frame_num": 4,
28
+ "max_object_pointers_in_encoder": 16,
29
+ "memory_attention_downsample_rate": 1,
30
+ "memory_attention_dropout": 0.1,
31
+ "memory_attention_feed_forward_hidden_act": "relu",
32
+ "memory_attention_feed_forward_hidden_size": 2048,
33
+ "memory_attention_hidden_size": 256,
34
+ "memory_attention_num_attention_heads": 1,
35
+ "memory_attention_num_layers": 4,
36
+ "memory_attention_rope_dropout": 0.1,
37
+ "memory_attention_rope_feat_sizes": [
38
+ 72,
39
+ 72
40
+ ],
41
+ "memory_attention_rope_theta": 10000,
42
+ "memory_encoder_hidden_size": 256,
43
+ "memory_encoder_output_channels": 64,
44
+ "memory_fuser_embed_dim": 256,
45
+ "memory_fuser_hidden_act": "gelu",
46
+ "memory_fuser_intermediate_dim": 1024,
47
+ "memory_fuser_kernel_size": 7,
48
+ "memory_fuser_layer_scale_init_value": 1e-06,
49
+ "memory_fuser_num_layers": 2,
50
+ "memory_fuser_padding": 3,
51
+ "model_type": "sam3_tracker",
52
+ "multimask_max_pt_num": 1,
53
+ "multimask_min_pt_num": 0,
54
+ "multimask_output_for_tracking": true,
55
+ "multimask_output_in_sam": true,
56
+ "num_maskmem": 7,
57
+ "prompt_encoder_config": {
58
+ "hidden_act": "gelu",
59
+ "hidden_size": 256,
60
+ "image_size": 1008,
61
+ "layer_norm_eps": 1e-06,
62
+ "mask_input_channels": 16,
63
+ "model_type": "",
64
+ "num_point_embeddings": 4,
65
+ "patch_size": 14,
66
+ "scale": 1
67
+ },
68
+ "sigmoid_bias_for_mem_enc": -10.0,
69
+ "sigmoid_scale_for_mem_enc": 20.0,
70
+ "transformers_version": "5.0.0.dev0",
71
+ "vision_config": {
72
+ "backbone_config": {
73
+ "_name_or_path": "",
74
+ "add_cross_attention": false,
75
+ "architectures": null,
76
+ "attention_dropout": 0.0,
77
+ "bad_words_ids": null,
78
+ "begin_suppress_tokens": null,
79
+ "bos_token_id": null,
80
+ "chunk_size_feed_forward": 0,
81
+ "cross_attention_hidden_size": null,
82
+ "decoder_start_token_id": null,
83
+ "diversity_penalty": 0.0,
84
+ "do_sample": false,
85
+ "dtype": null,
86
+ "early_stopping": false,
87
+ "encoder_no_repeat_ngram_size": 0,
88
+ "eos_token_id": null,
89
+ "exponential_decay_length_penalty": null,
90
+ "finetuning_task": null,
91
+ "forced_bos_token_id": null,
92
+ "forced_eos_token_id": null,
93
+ "global_attn_indexes": [
94
+ 7,
95
+ 15,
96
+ 23,
97
+ 31
98
+ ],
99
+ "hidden_act": "gelu",
100
+ "hidden_dropout": 0.0,
101
+ "hidden_size": 1024,
102
+ "id2label": {
103
+ "0": "LABEL_0",
104
+ "1": "LABEL_1"
105
+ },
106
+ "image_size": 1008,
107
+ "initializer_range": 0.02,
108
+ "intermediate_size": 4736,
109
+ "is_decoder": false,
110
+ "is_encoder_decoder": false,
111
+ "label2id": {
112
+ "LABEL_0": 0,
113
+ "LABEL_1": 1
114
+ },
115
+ "layer_norm_eps": 1e-06,
116
+ "layer_scale_init_value": null,
117
+ "length_penalty": 1.0,
118
+ "max_length": 20,
119
+ "min_length": 0,
120
+ "model_type": "sam3_vit_model",
121
+ "no_repeat_ngram_size": 0,
122
+ "num_attention_heads": 16,
123
+ "num_beam_groups": 1,
124
+ "num_beams": 1,
125
+ "num_channels": 3,
126
+ "num_hidden_layers": 32,
127
+ "num_return_sequences": 1,
128
+ "output_attentions": false,
129
+ "output_hidden_states": false,
130
+ "output_scores": false,
131
+ "pad_token_id": null,
132
+ "patch_size": 14,
133
+ "prefix": null,
134
+ "pretrain_image_size": 336,
135
+ "problem_type": null,
136
+ "qkv_bias": true,
137
+ "remove_invalid_values": false,
138
+ "repetition_penalty": 1.0,
139
+ "return_dict": true,
140
+ "return_dict_in_generate": false,
141
+ "rope_theta": 10000.0,
142
+ "sep_token_id": null,
143
+ "suppress_tokens": null,
144
+ "task_specific_params": null,
145
+ "temperature": 1.0,
146
+ "tie_encoder_decoder": false,
147
+ "tie_word_embeddings": true,
148
+ "tokenizer_class": null,
149
+ "top_k": 50,
150
+ "top_p": 1.0,
151
+ "typical_p": 1.0,
152
+ "window_size": 24
153
+ },
154
+ "backbone_feature_sizes": [
155
+ [
156
+ 288,
157
+ 288
158
+ ],
159
+ [
160
+ 144,
161
+ 144
162
+ ],
163
+ [
164
+ 72,
165
+ 72
166
+ ]
167
+ ],
168
+ "fpn_hidden_size": 256,
169
+ "fpn_kernel_size": 2,
170
+ "fpn_stride": 2,
171
+ "hidden_act": "gelu",
172
+ "initializer_range": 0.02,
173
+ "layer_norm_eps": 1e-06,
174
+ "model_type": "sam3_vision_model",
175
+ "num_feature_levels": 3,
176
+ "scale_factors": [
177
+ 4.0,
178
+ 2.0,
179
+ 1.0,
180
+ 0.5
181
+ ]
182
+ }
183
+ }