Hugging Face's logo

Datasets:
s-tian
/
VP2

Tags:
robotics
video prediction
License:
Dataset card Files Community
s-tian commited on
Commit
4727523
·
verified ·
1 Parent(s): 961a009

Upload 26 files

Browse files
Files changed (27) hide show
  1. .gitattributes +6 -0
  2. vp2_benchmark_data/pretrained_models/fitvid/rdall_base/config.json +1 -0
  3. vp2_benchmark_data/pretrained_models/fitvid/rdall_base/model_epoch33 +3 -0
  4. vp2_benchmark_data/pretrained_models/fitvid/rdall_lpips_1/config.json +1 -0
  5. vp2_benchmark_data/pretrained_models/fitvid/rdall_lpips_1/model_epoch33 +3 -0
  6. vp2_benchmark_data/pretrained_models/fitvid/rdall_lpips_10/config.json +1 -0
  7. vp2_benchmark_data/pretrained_models/fitvid/rdall_lpips_10/model_epoch33 +3 -0
  8. vp2_benchmark_data/pretrained_models/fitvid/rs5k_base/config.json +1 -0
  9. vp2_benchmark_data/pretrained_models/fitvid/rs5k_base/model_epoch33 +3 -0
  10. vp2_benchmark_data/pretrained_models/fitvid/rs5k_lpips_1/config.json +1 -0
  11. vp2_benchmark_data/pretrained_models/fitvid/rs5k_lpips_1/model_epoch33 +3 -0
  12. vp2_benchmark_data/pretrained_models/fitvid/rs5k_lpips_10/config.json +1 -0
  13. vp2_benchmark_data/pretrained_models/fitvid/rs5k_lpips_10/model_epoch33 +3 -0
  14. vp2_benchmark_data/pretrained_models/mcvd/rdall_base/checkpoint_330000.pt +3 -0
  15. vp2_benchmark_data/pretrained_models/mcvd/rdall_base/config.yml +130 -0
  16. vp2_benchmark_data/pretrained_models/mcvd/rs5k_base/checkpoint_360000.pt +3 -0
  17. vp2_benchmark_data/pretrained_models/mcvd/rs5k_base/config.yml +116 -0
  18. vp2_benchmark_data/pretrained_models/struct-vrnn/rdall_base/checkpoint-210000.pt +3 -0
  19. vp2_benchmark_data/pretrained_models/struct-vrnn/rdall_base/config.yaml +87 -0
  20. vp2_benchmark_data/pretrained_models/struct-vrnn/rs5k_base/checkpoint-210000.pt +3 -0
  21. vp2_benchmark_data/pretrained_models/struct-vrnn/rs5k_base/config.yaml +77 -0
  22. vp2_benchmark_data/pretrained_models/svg_prime/rdall_base/model_33_sd.pth +3 -0
  23. vp2_benchmark_data/pretrained_models/svg_prime/rdall_lpips_1/model_33_sd.pth +3 -0
  24. vp2_benchmark_data/pretrained_models/svg_prime/rdall_lpips_10/model_33_sd.pth +3 -0
  25. vp2_benchmark_data/pretrained_models/svg_prime/rs5k_base/model_33_sd.pth +3 -0
  26. vp2_benchmark_data/pretrained_models/svg_prime/rs5k_lpips_1/model_33_sd.pth +3 -0
  27. vp2_benchmark_data/pretrained_models/svg_prime/rs5k_lpips_10/model_33_sd.pth +3 -0
.gitattributes CHANGED
@@ -66,3 +66,9 @@ vp2_benchmark_data/robodesk_benchmark_tasks/robodesk_push_red/goals.hdf5 filter=
66
  vp2_benchmark_data/robodesk_benchmark_tasks/robodesk_upright_block_off_table/goals.hdf5 filter=lfs diff=lfs merge=lfs -text
67
  vp2_benchmark_data/robosuite_benchmark_tasks/combined/demo.hdf5 filter=lfs diff=lfs merge=lfs -text
68
  vp2_benchmark_data/robosuite_benchmark_tasks/combined/rendered_256.hdf5 filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
66
  vp2_benchmark_data/robodesk_benchmark_tasks/robodesk_upright_block_off_table/goals.hdf5 filter=lfs diff=lfs merge=lfs -text
67
  vp2_benchmark_data/robosuite_benchmark_tasks/combined/demo.hdf5 filter=lfs diff=lfs merge=lfs -text
68
  vp2_benchmark_data/robosuite_benchmark_tasks/combined/rendered_256.hdf5 filter=lfs diff=lfs merge=lfs -text
69
+ vp2_benchmark_data/pretrained_models/fitvid/rdall_base/model_epoch33 filter=lfs diff=lfs merge=lfs -text
70
+ vp2_benchmark_data/pretrained_models/fitvid/rdall_lpips_1/model_epoch33 filter=lfs diff=lfs merge=lfs -text
71
+ vp2_benchmark_data/pretrained_models/fitvid/rdall_lpips_10/model_epoch33 filter=lfs diff=lfs merge=lfs -text
72
+ vp2_benchmark_data/pretrained_models/fitvid/rs5k_base/model_epoch33 filter=lfs diff=lfs merge=lfs -text
73
+ vp2_benchmark_data/pretrained_models/fitvid/rs5k_lpips_1/model_epoch33 filter=lfs diff=lfs merge=lfs -text
74
+ vp2_benchmark_data/pretrained_models/fitvid/rs5k_lpips_10/model_epoch33 filter=lfs diff=lfs merge=lfs -text
vp2_benchmark_data/pretrained_models/fitvid/rdall_base/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_kwargs": {"stage_sizes": [2, 2, 2, 2], "z_dim": 10, "g_dim": 128, "rnn_size": 256, "num_base_filters": 64, "first_block_shape": [8, 8, 512], "skip_type": "residual", "n_past": 2, "action_conditioned": true, "action_size": 5, "expand_decoder": 4, "stochastic": true, "video_channels": 3, "lecun_initialization": false}, "beta": 0.0001, "multistep": false, "is_inference": false, "depth_predictor": null, "normal_predictor": null, "policy_networks": null, "loss_weights": {"kld": 0.0001, "rgb": 1.0, "depth": 0.0, "normal": 0.0, "policy": 0.0, "lpips": 0.0, "tv": 0.0, "segmented_object": 0.0}, "rgb_loss_type": "l2", "corr_wise": false}
vp2_benchmark_data/pretrained_models/fitvid/rdall_base/model_epoch33 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7613f818a4418089e8658237e8a35fc2d3126a69d700d9735813b8bd87bc3a21
3
+ size 1239991061
vp2_benchmark_data/pretrained_models/fitvid/rdall_lpips_1/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_kwargs": {"stage_sizes": [2, 2, 2, 2], "z_dim": 10, "g_dim": 128, "rnn_size": 256, "num_base_filters": 64, "first_block_shape": [8, 8, 512], "skip_type": "residual", "n_past": 2, "action_conditioned": true, "action_size": 5, "expand_decoder": 4, "stochastic": true, "video_channels": 3, "lecun_initialization": false}, "beta": 0.0001, "multistep": false, "is_inference": false, "depth_predictor": null, "normal_predictor": null, "policy_networks": null, "loss_weights": {"kld": 0.0001, "rgb": 1.0, "depth": 0.0, "normal": 0.0, "policy": 0.0, "lpips": 1.0, "tv": 0.0, "segmented_object": 0.0}, "rgb_loss_type": "l2", "corr_wise": false}
vp2_benchmark_data/pretrained_models/fitvid/rdall_lpips_1/model_epoch33 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b6b3cce235f82499e982436c2c0aaac6ccf507c22c8139f9681d63eefcb2bcf
3
+ size 1239991061
vp2_benchmark_data/pretrained_models/fitvid/rdall_lpips_10/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_kwargs": {"stage_sizes": [2, 2, 2, 2], "z_dim": 10, "g_dim": 128, "rnn_size": 256, "num_base_filters": 64, "first_block_shape": [8, 8, 512], "skip_type": "residual", "n_past": 2, "action_conditioned": true, "action_size": 5, "expand_decoder": 4, "stochastic": true, "video_channels": 3, "lecun_initialization": false}, "beta": 0.0001, "multistep": false, "is_inference": false, "depth_predictor": null, "normal_predictor": null, "policy_networks": null, "loss_weights": {"kld": 0.0001, "rgb": 1.0, "depth": 0.0, "normal": 0.0, "policy": 0.0, "lpips": 10.0, "tv": 0.0, "segmented_object": 0.0}, "rgb_loss_type": "l2", "corr_wise": false}
vp2_benchmark_data/pretrained_models/fitvid/rdall_lpips_10/model_epoch33 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d62292b1cf352845d5d495d67596d59539b9a21fd80e74cd7cc33f8b0928b59d
3
+ size 1239991061
vp2_benchmark_data/pretrained_models/fitvid/rs5k_base/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_kwargs": {"stage_sizes": [2, 2, 2, 2], "z_dim": 10, "g_dim": 128, "rnn_size": 256, "num_base_filters": 64, "first_block_shape": [8, 8, 512], "skip_type": "residual", "n_past": 2, "action_conditioned": true, "action_size": 4, "expand_decoder": 4, "stochastic": true, "video_channels": 3, "lecun_initialization": false}, "beta": 0.0001, "multistep": false, "is_inference": false, "depth_predictor": null, "normal_predictor": null, "policy_networks": null, "loss_weights": {"kld": 0.0001, "rgb": 1.0, "depth": 0.0, "normal": 0.0, "policy": 0.0, "lpips": 0.0, "tv": 0.0, "segmented_object": 0.0}, "rgb_loss_type": "l2", "corr_wise": false}
vp2_benchmark_data/pretrained_models/fitvid/rs5k_base/model_epoch33 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a364cabefef73e35ff324a821dd2e6f8a5a4b2068cf8d527e8b7ab8783d07f6a
3
+ size 1239990037
vp2_benchmark_data/pretrained_models/fitvid/rs5k_lpips_1/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_kwargs": {"stage_sizes": [2, 2, 2, 2], "z_dim": 10, "g_dim": 128, "rnn_size": 256, "num_base_filters": 64, "first_block_shape": [8, 8, 512], "skip_type": "residual", "n_past": 2, "action_conditioned": true, "action_size": 4, "expand_decoder": 4, "stochastic": true, "video_channels": 3, "lecun_initialization": false}, "beta": 0.0001, "multistep": false, "is_inference": false, "depth_predictor": null, "normal_predictor": null, "policy_networks": null, "loss_weights": {"kld": 0.0001, "rgb": 1.0, "depth": 0.0, "normal": 0.0, "policy": 0.0, "lpips": 1.0, "tv": 0.0, "segmented_object": 0.0}, "rgb_loss_type": "l2", "corr_wise": false}
vp2_benchmark_data/pretrained_models/fitvid/rs5k_lpips_1/model_epoch33 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7780112fd3b715d15d047406c2d7e5e3866b256f4749e1937cc7cbbe2cda6db9
3
+ size 1239990037
vp2_benchmark_data/pretrained_models/fitvid/rs5k_lpips_10/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_kwargs": {"stage_sizes": [2, 2, 2, 2], "z_dim": 10, "g_dim": 128, "rnn_size": 256, "num_base_filters": 64, "first_block_shape": [8, 8, 512], "skip_type": "residual", "n_past": 2, "action_conditioned": true, "action_size": 4, "expand_decoder": 4, "stochastic": true, "video_channels": 3, "lecun_initialization": false}, "beta": 0.0001, "multistep": false, "is_inference": false, "depth_predictor": null, "normal_predictor": null, "policy_networks": null, "loss_weights": {"kld": 0.0001, "rgb": 1.0, "depth": 0.0, "normal": 0.0, "policy": 0.0, "lpips": 10.0, "tv": 0.0, "segmented_object": 0.0}, "rgb_loss_type": "l2", "corr_wise": false}
vp2_benchmark_data/pretrained_models/fitvid/rs5k_lpips_10/model_epoch33 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:552ed90aa4c510a4a31b05da2c21fab76a8e5021785a0265faa3e4ba10174f7f
3
+ size 1239990037
vp2_benchmark_data/pretrained_models/mcvd/rdall_base/checkpoint_330000.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c66ada72db139a592f80c8579097c0abd512fcd83ebca4506d6b0c80656fee4d
3
+ size 897015077
vp2_benchmark_data/pretrained_models/mcvd/rdall_base/config.yml ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ data:
2
+ action_dimension: 5
3
+ cache_mode: low_dim
4
+ camera_view: camera
5
+ channels: 3
6
+ dataset: perceptual_metrics
7
+ dataset_files:
8
+ - /svl/u/stian/robodesk_open_drawer/noise_0.1_256.hdf5
9
+ - /svl/u/stian/robodesk_open_drawer/noise_0.2_256.hdf5
10
+ - /svl/u/stian/robodesk_open_slide/noise_0.1_256.hdf5
11
+ - /svl/u/stian/robodesk_open_slide/noise_0.2_256.hdf5
12
+ - /svl/u/stian/robodesk_open_slide/noise_0.5_256.hdf5
13
+ - /svl/u/stian/robodesk_push_blue/noise_0.1_256.hdf5
14
+ - /svl/u/stian/robodesk_push_blue/noise_0.2_256.hdf5
15
+ - /svl/u/stian/robodesk_push_green/noise_0.1_256.hdf5
16
+ - /svl/u/stian/robodesk_push_green/noise_0.2_256.hdf5
17
+ - /svl/u/stian/robodesk_push_red/noise_0.1_256.hdf5
18
+ - /svl/u/stian/robodesk_push_red/noise_0.2_256.hdf5
19
+ - /svl/u/stian/robodesk_flat_block_off_table/noise_0.1_256.hdf5
20
+ - /svl/u/stian/robodesk_flat_block_off_table/noise_0.2_256.hdf5
21
+ - /svl/u/stian/robodesk_upright_block_off_table/noise_0.1_256.hdf5
22
+ - /svl/u/stian/robodesk_upright_block_off_table/noise_0.2_256.hdf5
23
+ gaussian_dequantization: false
24
+ image_size: 64
25
+ logit_transform: false
26
+ num_frames: 5
27
+ num_frames_cond: 2
28
+ num_frames_future: 0
29
+ num_workers: 0
30
+ prob_mask_cond: 0.0
31
+ prob_mask_future: 0.0
32
+ prob_mask_sync: false
33
+ random_flip: true
34
+ rescaled: true
35
+ uniform_dequantization: false
36
+ fast_fid:
37
+ batch_size: 1000
38
+ begin_ckpt: 5000
39
+ end_ckpt: 300000
40
+ ensemble: false
41
+ freq: 5000
42
+ n_steps_each: 0
43
+ num_samples: 1000
44
+ pr_nn_k: 3
45
+ step_lr: 0.0
46
+ verbose: false
47
+ model:
48
+ arch: unetmore
49
+ attn_resolutions:
50
+ - 8
51
+ - 16
52
+ - 32
53
+ ch_mult:
54
+ - 1
55
+ - 2
56
+ - 3
57
+ - 4
58
+ cond_emb: false
59
+ conditional: true
60
+ depth: deep
61
+ dropout: 0.1
62
+ ema: true
63
+ ema_rate: 0.999
64
+ gamma: false
65
+ n_head_channels: 64
66
+ ngf: 64
67
+ noise_in_cond: false
68
+ nonlinearity: swish
69
+ normalization: InstanceNorm++
70
+ num_classes: 1000
71
+ num_res_blocks: 2
72
+ output_all_frames: false
73
+ sigma_begin: 0.02
74
+ sigma_dist: linear
75
+ sigma_end: 0.0001
76
+ spade: true
77
+ spade_dim: 128
78
+ spec_norm: false
79
+ time_conditional: true
80
+ type: v1
81
+ version: DDPM
82
+ optim:
83
+ amsgrad: false
84
+ beta1: 0.9
85
+ eps: 1.0e-08
86
+ grad_clip: 1.0
87
+ lr: 0.0002
88
+ optimizer: Adam
89
+ warmup: 1000
90
+ weight_decay: 0.0
91
+ sampling:
92
+ batch_size: 64
93
+ ckpt_id: 0
94
+ clip_before: true
95
+ consistent: true
96
+ data_init: false
97
+ denoise: true
98
+ fid: false
99
+ final_only: true
100
+ fvd: true
101
+ init_prev_t: -1.0
102
+ inpainting: false
103
+ interpolation: false
104
+ max_data_iter: 100000
105
+ n_interpolations: 15
106
+ n_steps_each: 0
107
+ num_frames_pred: 10
108
+ num_samples4fid: 10000
109
+ num_samples4fvd: 10000
110
+ one_frame_at_a_time: false
111
+ preds_per_test: 1
112
+ ssim: true
113
+ step_lr: 0.0
114
+ subsample: 100
115
+ train: true
116
+ test:
117
+ batch_size: 100
118
+ begin_ckpt: 5000
119
+ end_ckpt: 300000
120
+ training:
121
+ L1: false
122
+ batch_size: 16
123
+ log_all_sigmas: false
124
+ log_freq: 50
125
+ n_epochs: 1000000
126
+ n_iters: 3000001
127
+ sample_freq: 15000
128
+ snapshot_freq: 10
129
+ snapshot_sampling: true
130
+ val_freq: 100
vp2_benchmark_data/pretrained_models/mcvd/rs5k_base/checkpoint_360000.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d871847d01e2fe3cbb9a8da20d93934c7275b253ee8a2ce0d74a767a5293e192
3
+ size 890711333
vp2_benchmark_data/pretrained_models/mcvd/rs5k_base/config.yml ADDED
@@ -0,0 +1,116 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ data:
2
+ action_dimension: 4
3
+ cache_mode: all_nogetitem
4
+ camera_view: agentview_shift_2
5
+ channels: 3
6
+ dataset: perceptual_metrics
7
+ dataset_files:
8
+ - /svl/u/stian/robosuite_50k_1/5k_slice_rendered_256.hdf5
9
+ gaussian_dequantization: false
10
+ image_size: 64
11
+ logit_transform: false
12
+ num_frames: 5
13
+ num_frames_cond: 2
14
+ num_frames_future: 0
15
+ num_workers: 0
16
+ prob_mask_cond: 0.0
17
+ prob_mask_future: 0.0
18
+ prob_mask_sync: false
19
+ random_flip: true
20
+ rescaled: true
21
+ uniform_dequantization: false
22
+ fast_fid:
23
+ batch_size: 1000
24
+ begin_ckpt: 5000
25
+ end_ckpt: 300000
26
+ ensemble: false
27
+ freq: 5000
28
+ n_steps_each: 0
29
+ num_samples: 1000
30
+ pr_nn_k: 3
31
+ step_lr: 0.0
32
+ verbose: false
33
+ model:
34
+ arch: unetmore
35
+ attn_resolutions:
36
+ - 8
37
+ - 16
38
+ - 32
39
+ ch_mult:
40
+ - 1
41
+ - 2
42
+ - 3
43
+ - 4
44
+ cond_emb: false
45
+ conditional: true
46
+ depth: deep
47
+ dropout: 0.1
48
+ ema: true
49
+ ema_rate: 0.999
50
+ gamma: false
51
+ n_head_channels: 64
52
+ ngf: 64
53
+ noise_in_cond: false
54
+ nonlinearity: swish
55
+ normalization: InstanceNorm++
56
+ num_classes: 1000
57
+ num_res_blocks: 2
58
+ output_all_frames: false
59
+ sigma_begin: 0.02
60
+ sigma_dist: linear
61
+ sigma_end: 0.0001
62
+ spade: true
63
+ spade_dim: 128
64
+ spec_norm: false
65
+ time_conditional: true
66
+ type: v1
67
+ version: DDPM
68
+ optim:
69
+ amsgrad: false
70
+ beta1: 0.9
71
+ eps: 1.0e-08
72
+ grad_clip: 1.0
73
+ lr: 0.0002
74
+ optimizer: Adam
75
+ warmup: 1000
76
+ weight_decay: 0.0
77
+ sampling:
78
+ batch_size: 64
79
+ ckpt_id: 0
80
+ clip_before: true
81
+ consistent: true
82
+ data_init: false
83
+ denoise: true
84
+ fid: false
85
+ final_only: true
86
+ fvd: true
87
+ init_prev_t: -1.0
88
+ inpainting: false
89
+ interpolation: false
90
+ max_data_iter: 100000
91
+ n_interpolations: 15
92
+ n_steps_each: 0
93
+ num_frames_pred: 10
94
+ num_samples4fid: 10000
95
+ num_samples4fvd: 10000
96
+ one_frame_at_a_time: false
97
+ preds_per_test: 1
98
+ ssim: true
99
+ step_lr: 0.0
100
+ subsample: 100
101
+ train: true
102
+ test:
103
+ batch_size: 100
104
+ begin_ckpt: 5000
105
+ end_ckpt: 300000
106
+ training:
107
+ L1: false
108
+ batch_size: 64
109
+ log_all_sigmas: false
110
+ log_freq: 50
111
+ n_epochs: 1000000
112
+ n_iters: 3000001
113
+ sample_freq: 15000
114
+ snapshot_freq: 15000
115
+ snapshot_sampling: true
116
+ val_freq: 100
vp2_benchmark_data/pretrained_models/struct-vrnn/rdall_base/checkpoint-210000.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bf50cce15f89a824fde33e0612487eede0783fb2ea23edc04d896b917ba08ad5
3
+ size 204401183
vp2_benchmark_data/pretrained_models/struct-vrnn/rdall_base/config.yaml ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ dataset:
2
+ num_steps: 12
3
+ camera_name: camera
4
+ cache_mode: low_dim
5
+ dataset_files:
6
+ - /svl/u/stian/robodesk_open_drawer/noise_0.1_256.hdf5
7
+ - /svl/u/stian/robodesk_open_drawer/noise_0.2_256.hdf5
8
+ - /svl/u/stian/robodesk_open_slide/noise_0.1_256.hdf5
9
+ - /svl/u/stian/robodesk_open_slide/noise_0.2_256.hdf5
10
+ - /svl/u/stian/robodesk_open_slide/noise_0.5_256.hdf5
11
+ - /svl/u/stian/robodesk_push_blue/noise_0.1_256.hdf5
12
+ - /svl/u/stian/robodesk_push_blue/noise_0.2_256.hdf5
13
+ - /svl/u/stian/robodesk_push_green/noise_0.1_256.hdf5
14
+ - /svl/u/stian/robodesk_push_green/noise_0.2_256.hdf5
15
+ - /svl/u/stian/robodesk_push_red/noise_0.1_256.hdf5
16
+ - /svl/u/stian/robodesk_push_red/noise_0.2_256.hdf5
17
+ - /svl/u/stian/robodesk_flat_block_off_table/noise_0.1_256.hdf5
18
+ - /svl/u/stian/robodesk_flat_block_off_table/noise_0.2_256.hdf5
19
+ - /svl/u/stian/robodesk_upright_block_off_table/noise_0.1_256.hdf5
20
+ - /svl/u/stian/robodesk_upright_block_off_table/noise_0.2_256.hdf5
21
+ data_type: robomimic
22
+ run_name: robodesk
23
+ optimizer:
24
+ _target_: torch.optim.Adam
25
+ lr: 0.0003
26
+ batch_size: 8
27
+ train_steps: 500000
28
+ train_epoch_length: 3000
29
+ log_every: 30000
30
+ log_images_every: 2000
31
+ num_videos_to_log: 6
32
+ device: cuda
33
+ num_workers: 8
34
+ training_weights:
35
+ reconstruction_loss: 10.0
36
+ kld_loss: 0.0001
37
+ temporal_separation_loss: 0.0
38
+ coord_pred_loss: 1.0
39
+ sparsity_loss: 0
40
+ model:
41
+ _target_: struct_vrnn.model.keypoint_vrnn.StructVRNN
42
+ temporal_separation_loss_sigma: 0.02
43
+ keypoint_detector:
44
+ _target_: struct_vrnn.model.keypoint_detector.KeypointDetector
45
+ num_keypoints: 64
46
+ keypoint_sigma: 3.0
47
+ encoder:
48
+ _target_: struct_vrnn.model.keypoint_detector.ImageEncoder
49
+ num_input_channels: 5
50
+ input_map_size: 64
51
+ output_map_size: 16
52
+ initial_num_filters: 32
53
+ layers_per_scale: 2
54
+ appearance_encoder:
55
+ _target_: struct_vrnn.model.keypoint_detector.ImageEncoder
56
+ num_input_channels: 3
57
+ input_map_size: 64
58
+ output_map_size: 16
59
+ initial_num_filters: 32
60
+ layers_per_scale: 2
61
+ decoder:
62
+ _target_: struct_vrnn.model.keypoint_detector.ImageDecoder
63
+ num_input_channels: 258
64
+ input_map_size: 16
65
+ output_map_size: 64
66
+ initial_num_filters: 256
67
+ layers_per_scale: 3
68
+ dynamics:
69
+ _target_: struct_vrnn.model.dynamics.Dynamics
70
+ prior:
71
+ _target_: struct_vrnn.model.lstm.MultiGaussianLSTM
72
+ input_size: 192
73
+ output_size: 16
74
+ hidden_size: 512
75
+ num_layers: 2
76
+ posterior:
77
+ _target_: struct_vrnn.model.lstm.MultiGaussianLSTM
78
+ input_size: 192
79
+ output_size: 16
80
+ hidden_size: 512
81
+ num_layers: 2
82
+ dynamics:
83
+ _target_: struct_vrnn.model.lstm.MultiGaussianLSTM
84
+ input_size: 213
85
+ output_size: 192
86
+ hidden_size: 512
87
+ num_layers: 2
vp2_benchmark_data/pretrained_models/struct-vrnn/rs5k_base/checkpoint-210000.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:05e1392aa431f9b4176c34c0264aa6a1263c25c2727fca8579a758d09ee7a4ae
3
+ size 204395039
vp2_benchmark_data/pretrained_models/struct-vrnn/rs5k_base/config.yaml ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ dataset:
2
+ num_steps: 12
3
+ camera_name: agentview_shift_2
4
+ cache_mode: low_dim
5
+ dataset_files:
6
+ - /svl/u/stian/robosuite_50k_1/rendered_256.hdf5
7
+ - /svl/u/stian/robosuite_50k_2/rendered_256.hdf5
8
+ - /svl/u/stian/robosuite_50k_3/rendered_256.hdf5
9
+ - /svl/u/stian/robosuite_50k_4/rendered_256.hdf5
10
+ - /svl/u/stian/robosuite_50k_5/rendered_256.hdf5
11
+ data_type: robomimic
12
+ run_name: robosuite
13
+ optimizer:
14
+ _target_: torch.optim.Adam
15
+ lr: 0.0003
16
+ batch_size: 8
17
+ train_steps: 500000
18
+ train_epoch_length: 3000
19
+ log_every: 30000
20
+ log_images_every: 2000
21
+ num_videos_to_log: 6
22
+ device: cuda
23
+ num_workers: 8
24
+ training_weights:
25
+ reconstruction_loss: 10.0
26
+ kld_loss: 0.0001
27
+ temporal_separation_loss: 0.0
28
+ coord_pred_loss: 1.0
29
+ sparsity_loss: 0
30
+ model:
31
+ _target_: struct_vrnn.model.keypoint_vrnn.StructVRNN
32
+ temporal_separation_loss_sigma: 0.02
33
+ keypoint_detector:
34
+ _target_: struct_vrnn.model.keypoint_detector.KeypointDetector
35
+ num_keypoints: 64
36
+ keypoint_sigma: 3.0
37
+ encoder:
38
+ _target_: struct_vrnn.model.keypoint_detector.ImageEncoder
39
+ num_input_channels: 5
40
+ input_map_size: 64
41
+ output_map_size: 16
42
+ initial_num_filters: 32
43
+ layers_per_scale: 2
44
+ appearance_encoder:
45
+ _target_: struct_vrnn.model.keypoint_detector.ImageEncoder
46
+ num_input_channels: 3
47
+ input_map_size: 64
48
+ output_map_size: 16
49
+ initial_num_filters: 32
50
+ layers_per_scale: 2
51
+ decoder:
52
+ _target_: struct_vrnn.model.keypoint_detector.ImageDecoder
53
+ num_input_channels: 258
54
+ input_map_size: 16
55
+ output_map_size: 64
56
+ initial_num_filters: 256
57
+ layers_per_scale: 3
58
+ dynamics:
59
+ _target_: struct_vrnn.model.dynamics.Dynamics
60
+ prior:
61
+ _target_: struct_vrnn.model.lstm.MultiGaussianLSTM
62
+ input_size: 192
63
+ output_size: 16
64
+ hidden_size: 512
65
+ num_layers: 2
66
+ posterior:
67
+ _target_: struct_vrnn.model.lstm.MultiGaussianLSTM
68
+ input_size: 192
69
+ output_size: 16
70
+ hidden_size: 512
71
+ num_layers: 2
72
+ dynamics:
73
+ _target_: struct_vrnn.model.lstm.MultiGaussianLSTM
74
+ input_size: 212
75
+ output_size: 192
76
+ hidden_size: 512
77
+ num_layers: 2
vp2_benchmark_data/pretrained_models/svg_prime/rdall_base/model_33_sd.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c870a90e4dab5a3389560e29b84e4124426c89281ed1c9fb781502389c0094dc
3
+ size 995619143
vp2_benchmark_data/pretrained_models/svg_prime/rdall_lpips_1/model_33_sd.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6c3162b13964d4fb6c62d579b38c0c644b4ee1d700bdd39bdb67b78459bea7eb
3
+ size 995619143
vp2_benchmark_data/pretrained_models/svg_prime/rdall_lpips_10/model_33_sd.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c95354197696d1cee0b5700805c1fb7195504173f36466e6755ae1081d642edb
3
+ size 995619143
vp2_benchmark_data/pretrained_models/svg_prime/rs5k_base/model_33_sd.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:00a16a92e243fb79ac0a5f03ddcba33840c0ee82ab8bfc4490bd198e63778fa0
3
+ size 995563015
vp2_benchmark_data/pretrained_models/svg_prime/rs5k_lpips_1/model_33_sd.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ed8a2ddc974519e47a640ca387f644798abbb69376eb8310e33e145f02390dc6
3
+ size 995563015
vp2_benchmark_data/pretrained_models/svg_prime/rs5k_lpips_10/model_33_sd.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:695790bf2ab42ef809ac5d52f1eba99a7c0554fc3fdc28f7dc802bdab9672593
3
+ size 995563015