Hugging Face's logo

iantc104
/
gaze_real_occluded_insertion_dinov2_baseline

Robotics
Safetensors
lerobot
act
model_hub_mixin
pytorch_model_hub_mixin
Model card Files Community
iantc104 commited on
Commit
e641a40
·
verified ·
1 Parent(s): 9a4e918

Upload YAML configuration

Browse files
Files changed (1) hide show
  1. config.yaml +187 -0
config.yaml ADDED
@@ -0,0 +1,187 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ resume: false
2
+ device: cuda
3
+ use_amp: false
4
+ seed: 0
5
+ dataset_repo_id: iantc104/gaze_real_occluded_insertion
6
+ video_backend: pyav
7
+ training:
8
+ offline_steps: 100000
9
+ num_workers: 4
10
+ batch_size: 8
11
+ eval_freq: -1
12
+ log_freq: 100
13
+ save_checkpoint: true
14
+ save_freq: 10000
15
+ online_steps: 0
16
+ online_rollout_n_episodes: 1
17
+ online_rollout_batch_size: 1
18
+ online_steps_between_rollouts: 1
19
+ online_sampling_ratio: 0.5
20
+ online_env_seed: null
21
+ online_buffer_capacity: null
22
+ online_buffer_seed_size: 0
23
+ do_online_rollout_async: false
24
+ image_transforms:
25
+ enable: false
26
+ max_num_transforms: 3
27
+ random_order: false
28
+ brightness:
29
+ weight: 1
30
+ min_max:
31
+ - 0.8
32
+ - 1.2
33
+ contrast:
34
+ weight: 1
35
+ min_max:
36
+ - 0.8
37
+ - 1.2
38
+ saturation:
39
+ weight: 1
40
+ min_max:
41
+ - 0.5
42
+ - 1.5
43
+ hue:
44
+ weight: 1
45
+ min_max:
46
+ - -0.05
47
+ - 0.05
48
+ sharpness:
49
+ weight: 1
50
+ min_max:
51
+ - 0.8
52
+ - 1.2
53
+ lr: 1.0e-05
54
+ lr_backbone: 1.0e-05
55
+ weight_decay: 0.0001
56
+ grad_clip_norm: 10
57
+ delta_timestamps:
58
+ action:
59
+ - 0.0
60
+ - 0.030303030303030304
61
+ - 0.06060606060606061
62
+ - 0.09090909090909091
63
+ - 0.12121212121212122
64
+ - 0.15151515151515152
65
+ - 0.18181818181818182
66
+ - 0.21212121212121213
67
+ - 0.24242424242424243
68
+ - 0.2727272727272727
69
+ - 0.30303030303030304
70
+ - 0.3333333333333333
71
+ - 0.36363636363636365
72
+ - 0.3939393939393939
73
+ - 0.42424242424242425
74
+ - 0.45454545454545453
75
+ - 0.48484848484848486
76
+ - 0.5151515151515151
77
+ - 0.5454545454545454
78
+ - 0.5757575757575758
79
+ - 0.6060606060606061
80
+ - 0.6363636363636364
81
+ - 0.6666666666666666
82
+ - 0.696969696969697
83
+ - 0.7272727272727273
84
+ - 0.7575757575757576
85
+ - 0.7878787878787878
86
+ - 0.8181818181818182
87
+ - 0.8484848484848485
88
+ - 0.8787878787878788
89
+ - 0.9090909090909091
90
+ - 0.9393939393939394
91
+ - 0.9696969696969697
92
+ - 1.0
93
+ - 1.0303030303030303
94
+ - 1.0606060606060606
95
+ - 1.0909090909090908
96
+ - 1.121212121212121
97
+ - 1.1515151515151516
98
+ - 1.1818181818181819
99
+ - 1.2121212121212122
100
+ - 1.2424242424242424
101
+ - 1.2727272727272727
102
+ - 1.303030303030303
103
+ - 1.3333333333333333
104
+ - 1.3636363636363635
105
+ - 1.393939393939394
106
+ - 1.4242424242424243
107
+ - 1.4545454545454546
108
+ - 1.4848484848484849
109
+ eval:
110
+ n_episodes: 50
111
+ batch_size: 10
112
+ use_async_envs: false
113
+ wandb:
114
+ enable: true
115
+ disable_artifact: true
116
+ project: real_occluded_insertion
117
+ notes: ''
118
+ fps: 33
119
+ env:
120
+ name: real_world
121
+ task: null
122
+ state_dim: 21
123
+ action_dim: 21
124
+ fps: ${fps}
125
+ override_dataset_stats:
126
+ observation.images.left_eye_cam:
127
+ mean:
128
+ - - - 0.485
129
+ - - - 0.456
130
+ - - - 0.406
131
+ std:
132
+ - - - 0.229
133
+ - - - 0.224
134
+ - - - 0.225
135
+ observation.images.right_eye_cam:
136
+ mean:
137
+ - - - 0.485
138
+ - - - 0.456
139
+ - - - 0.406
140
+ std:
141
+ - - - 0.229
142
+ - - - 0.224
143
+ - - - 0.225
144
+ policy:
145
+ name: act
146
+ n_obs_steps: 1
147
+ chunk_size: 50
148
+ n_action_steps: 50
149
+ input_shapes:
150
+ observation.images.left_eye_cam:
151
+ - 3
152
+ - 480
153
+ - 640
154
+ observation.images.right_eye_cam:
155
+ - 3
156
+ - 480
157
+ - 640
158
+ observation.state:
159
+ - ${env.state_dim}
160
+ output_shapes:
161
+ action:
162
+ - ${env.action_dim}
163
+ input_normalization_modes:
164
+ observation.images.left_eye_cam: mean_std
165
+ observation.images.right_eye_cam: mean_std
166
+ observation.state: mean_std
167
+ output_normalization_modes:
168
+ action: mean_std
169
+ image_size:
170
+ - 336
171
+ - 448
172
+ vision_backbone: dinov2
173
+ pretrained_backbone_weights: dinov2_vits14_reg
174
+ freeze_backbone: true
175
+ pre_norm: false
176
+ dim_model: 512
177
+ n_heads: 8
178
+ dim_feedforward: 3200
179
+ feedforward_activation: relu
180
+ n_encoder_layers: 4
181
+ n_decoder_layers: 1
182
+ use_vae: true
183
+ latent_dim: 32
184
+ n_vae_encoder_layers: 4
185
+ temporal_ensemble_coeff: null
186
+ dropout: 0.1
187
+ kl_weight: 10.0