Hugging Face's logo

ChenyuanC
/
3D-CovDiffusion

Robotics
PyTorch
Safetensors
diffusion-policy
3d-point-cloud
coverage-path-planning
Model card Files
xet
 Community
ChenyuanC commited on
Commit
c1f89f2
·
verified ·
1 Parent(s): 2f26186

Add files using upload-large-folder tool

Browse files
Files changed (15) hide show
  1. README.md +103 -0
  2. SHA256SUMS +14 -0
  3. containers/config.yaml +61 -0
  4. containers/metrics.json +31 -0
  5. containers/model.safetensors +3 -0
  6. cuboids/config.yaml +53 -0
  7. cuboids/metrics.json +31 -0
  8. cuboids/model.safetensors +3 -0
  9. manifest.json +186 -0
  10. shelves/config.yaml +53 -0
  11. shelves/metrics.json +31 -0
  12. shelves/model.safetensors +3 -0
  13. windows/config.yaml +61 -0
  14. windows/metrics.json +31 -0
  15. windows/model.safetensors +3 -0
README.md ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: pytorch
3
+ tags:
4
+ - robotics
5
+ - diffusion-policy
6
+ - 3d-point-cloud
7
+ - coverage-path-planning
8
+ - safetensors
9
+ ---
10
+
11
+ # 3D-CovDiffusion checkpoints
12
+
13
+ Category-specific pretrained policies for **3D-CovDiffusion: 3D-Aware
14
+ Diffusion Policy for Coverage Path Planning**.
15
+
16
+ - Project page: https://crystalccy1.github.io/3D-CovDiffusion/
17
+ - Code: https://github.com/crystalccy1/3D-CovDiffusion
18
+ - Train-ready dataset: https://huggingface.co/datasets/ChenyuanC/3D-CovDiffusion-Train-Ready
19
+
20
+ ## Released models
21
+
22
+ Each directory contains the EMA weights selected by the original seed-42
23
+ training run, the exact inference configuration, and provenance metadata.
24
+
25
+ | Directory | Dataset | Training run | Top-k selection PCD ↓ |
26
+ |:--|:--|:--|--:|
27
+ | `windows/` | `windows-v2` | `TML4Q-S42` | 10.410878 |
28
+ | `cuboids/` | `cuboids-v2` | `X1PD1-S42` | 6.612324 |
29
+ | `shelves/` | `shelves-v2` | `52VCU-S42` | 10.069027 |
30
+ | `containers/` | `containers-v2` | `ODAV4-S42` | 347.932620 |
31
+
32
+ The values above are the per-run validation monitor used for top-k checkpoint
33
+ selection. They are not the three-seed test results reported in the paper.
34
+ Containers is a separate low-data experiment; `ODAV4-S42` is released because
35
+ it is the checkpoint referenced by the original in-domain, OOD, and video
36
+ evaluation scripts.
37
+
38
+ ## Download and evaluate
39
+
40
+ Install the Hugging Face CLI and download one category:
41
+
42
+ ```bash
43
+ hf download ChenyuanC/3D-CovDiffusion \
44
+ --include "windows/*" \
45
+ --local-dir checkpoints/3d-covdiffusion
46
+ ```
47
+
48
+ With the public code and evaluation data configured:
49
+
50
+ ```bash
51
+ CUDA_VISIBLE_DEVICES=0 PYTHONPATH=. python evaluate.py \
52
+ --checkpoint_path checkpoints/3d-covdiffusion/windows/model.safetensors \
53
+ --config_path checkpoints/3d-covdiffusion/windows/config.yaml \
54
+ --dataset_name windows-v2 \
55
+ --dataset_split test \
56
+ --eval_episodes 20 \
57
+ --workers 4
58
+ ```
59
+
60
+ Replace `windows` / `windows-v2` with `cuboids`, `shelves`, or `containers`.
61
+ The adjacent `config.yaml` is discovered automatically, so `--config_path` can
62
+ be omitted when the original directory layout is preserved.
63
+
64
+ ## Format and integrity
65
+
66
+ - `model.safetensors` contains the complete EMA policy state, including the
67
+ action and point-cloud normalizer tensors.
68
+ - Optimizer state, raw non-EMA weights, Python/Dill pickles, W&B metadata, and
69
+ machine-local paths are intentionally excluded.
70
+ - `metrics.json` records the source run, epoch, step, selection metric, source
71
+ checkpoint digest, release digest, and exact tensor-roundtrip validation.
72
+ - `manifest.json` and `SHA256SUMS` provide repository-wide integrity metadata.
73
+
74
+ The original training checkpoints must be treated as trusted pickle files. The
75
+ files in this repository use the tensor-only safetensors format and are the
76
+ recommended artifacts for inference.
77
+
78
+ ## Intended use and limitations
79
+
80
+ These checkpoints generate ordered 6-DoF coverage-trajectory chunks from a
81
+ 5,120-point observation and recent execution history. They are research
82
+ artifacts evaluated on the corresponding geometry categories; they are not a
83
+ certified motion-planning or robot-safety system. Validate collision handling,
84
+ kinematic feasibility, workcell constraints, and emergency behavior before any
85
+ physical deployment.
86
+
87
+ ## License
88
+
89
+ No standalone repository or model-weight license has been selected yet.
90
+ Third-party components remain subject to their original terms; see the notices
91
+ in the code repository.
92
+
93
+ ## Citation
94
+
95
+ ```bibtex
96
+ @misc{chen2026_3dcovdiffusion,
97
+ title = {{3D-CovDiffusion}: 3D-Aware Diffusion Policy for Coverage Path Planning},
98
+ author = {Chen, Chenyuan and Ding, Haoran and Ding, Ran and Liu, Tianyu
99
+ and He, Zewen and Duan, Anqing and Nakamura, Yoshihiko},
100
+ year = {2026},
101
+ note = {Manuscript}
102
+ }
103
+ ```
SHA256SUMS ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 829a9c0e7c0e93ad101ab7bf8a5d6a587e80d4799384b02cdde62d52fa83b948 README.md
2
+ 921479d3af21384ecc0362c9516b019df209f593a1b3ceed985f4d85482e42cc containers/config.yaml
3
+ 7945015f5b38b13551ec707cfa36b39aa1b62233afa9265471c9c864362f04fb containers/metrics.json
4
+ 19559bfecdba3375d8ebe6fc043a53c10297cedfa226023b484bc2fad8f30102 containers/model.safetensors
5
+ a405bcbaa8bd80a73353d786e37972ff93f91d5e075aa171d443adec45e3cc4d cuboids/config.yaml
6
+ 7587101b368c5c0de556aa93c1a2e36767428bd073e0f21128733ead1c6496c1 cuboids/metrics.json
7
+ 0116ddc97f31055d8becf3b6ce89d00d66bbfa6b27bf3cdb8ca8fcd525f0c76d cuboids/model.safetensors
8
+ c19e9403f130432d1599cfc98b3eeb2a4a819fabf9e46536a0ad34c7700dc935 manifest.json
9
+ a33f9643c53f85d7d6f55a105cf19e6844d194e3bd6535c36701a37c5b002395 shelves/config.yaml
10
+ c2473be1acf7b900db676b87fdbe73688b864523451b1ac246ee5d1a1794dcc6 shelves/metrics.json
11
+ aa84b6c73a37bb3906df9b60cf69c4942999f0520ef85977e91b1120338e3256 shelves/model.safetensors
12
+ b328bc6090444ed4655f64e5d40eade119fb1f9a8f5269195c77484252648c00 windows/config.yaml
13
+ 3ddd8f5049bb7312a5416c5684878524450271e59a4b728af227df2d3b775a27 windows/metrics.json
14
+ 2ab8dc83ce59b703d37cd269964786c5c61a31c3741e434b90b545d498909d5e windows/model.safetensors
containers/config.yaml ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ format_version: 3dcov-inference-config-v1
2
+ task_name: CovDiffusion
3
+ io_type: CovDiffusion
4
+ dataset:
5
+ - containers-v2
6
+ action_dim: 24
7
+ horizon: 16
8
+ n_action_steps: 100
9
+ n_obs_steps: 1
10
+ encoder_output_dim: 256
11
+ shape_meta:
12
+ obs:
13
+ point_cloud:
14
+ shape:
15
+ - 5120
16
+ - 3
17
+ low_dim:
18
+ shape:
19
+ - 24
20
+ action:
21
+ shape:
22
+ - 24
23
+ diffusion:
24
+ model_type: dp3
25
+ diffusion_step_embed_dim: 128
26
+ down_dims:
27
+ - 512
28
+ - 1024
29
+ - 2048
30
+ kernel_size: 5
31
+ n_groups: 8
32
+ condition_type: film
33
+ use_down_condition: true
34
+ use_mid_condition: true
35
+ use_up_condition: true
36
+ num_inference_steps: 10
37
+ obs_as_global_cond: true
38
+ model:
39
+ backbone: dp3
40
+ affinetrans: false
41
+ hidden_size:
42
+ - 1024
43
+ - 1024
44
+ pretrained: false
45
+ pretrained_custom: null
46
+ load_strict: false
47
+ noise_scheduler:
48
+ _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
49
+ num_train_timesteps: 100
50
+ beta_start: 0.0001
51
+ beta_end: 0.02
52
+ beta_schedule: squaredcos_cap_v2
53
+ clip_sample: true
54
+ set_alpha_to_one: true
55
+ steps_offset: 0
56
+ prediction_type: sample
57
+ partial_observation_enabled: false
58
+ partial_observation_method: fixed_camera
59
+ partial_observation_ratio: 0.3
60
+ ablation_prev_traj: normal
61
+ random_traj_seed: 123
containers/metrics.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "category": "containers",
3
+ "dataset": "containers-v2",
4
+ "epoch": 500,
5
+ "format_version": "3dcov-checkpoint-metadata-v1",
6
+ "global_step": 141783,
7
+ "release": {
8
+ "exact_tensor_roundtrip": true,
9
+ "git_revision": "10403c5d7365d765522f279d892b581d1d4083e5",
10
+ "model_sha256": "19559bfecdba3375d8ebe6fc043a53c10297cedfa226023b484bc2fad8f30102",
11
+ "normalizer_fields": [
12
+ "action",
13
+ "point_cloud"
14
+ ],
15
+ "tensor_bytes": 1020966632,
16
+ "tensor_count": 182
17
+ },
18
+ "run_id": "ODAV4-S42",
19
+ "seed": 42,
20
+ "selection_metric": {
21
+ "mode": "min",
22
+ "name": "pred_cond_Pred_Cond_mean_point-wise_chamfer_distance",
23
+ "scope": "seed-42 validation/top-k selection",
24
+ "value": 347.93262
25
+ },
26
+ "source": {
27
+ "checkpoint_basename": "epoch=500-step=141783-pred_cond_Pred_Cond_mean_point-wise_chamfer_distance=347p9326.ckpt",
28
+ "sha256": "7c66fbf944afa0182e2863b5a163beb09a5696c29333143cf8f25594c5ffbfb8"
29
+ },
30
+ "weight_variant": "ema"
31
+ }
containers/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:19559bfecdba3375d8ebe6fc043a53c10297cedfa226023b484bc2fad8f30102
3
+ size 1020987472
cuboids/config.yaml ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ format_version: 3dcov-inference-config-v1
2
+ task_name: CovDiffusion
3
+ io_type: CovDiffusion
4
+ dataset:
5
+ - cuboids-v2
6
+ action_dim: 24
7
+ horizon: 16
8
+ n_action_steps: 100
9
+ n_obs_steps: 1
10
+ encoder_output_dim: 256
11
+ diffusion:
12
+ model_type: dp3
13
+ diffusion_step_embed_dim: 128
14
+ down_dims:
15
+ - 512
16
+ - 1024
17
+ - 2048
18
+ kernel_size: 5
19
+ n_groups: 8
20
+ condition_type: film
21
+ use_down_condition: true
22
+ use_mid_condition: true
23
+ use_up_condition: true
24
+ num_inference_steps: 10
25
+ obs_as_global_cond: true
26
+ model:
27
+ backbone: dp3
28
+ affinetrans: false
29
+ hidden_size:
30
+ - 1024
31
+ - 1024
32
+ pretrained: false
33
+ pretrained_custom: null
34
+ load_strict: false
35
+ noise_scheduler:
36
+ _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
37
+ num_train_timesteps: 100
38
+ beta_start: 0.0001
39
+ beta_end: 0.02
40
+ beta_schedule: squaredcos_cap_v2
41
+ clip_sample: true
42
+ set_alpha_to_one: true
43
+ steps_offset: 0
44
+ prediction_type: sample
45
+ shape_meta:
46
+ obs:
47
+ point_cloud:
48
+ shape:
49
+ - 5120
50
+ - 3
51
+ action:
52
+ shape:
53
+ - 24
cuboids/metrics.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "category": "cuboids",
3
+ "dataset": "cuboids-v2",
4
+ "epoch": 40,
5
+ "format_version": "3dcov-checkpoint-metadata-v1",
6
+ "global_step": 38007,
7
+ "release": {
8
+ "exact_tensor_roundtrip": true,
9
+ "git_revision": "10403c5d7365d765522f279d892b581d1d4083e5",
10
+ "model_sha256": "0116ddc97f31055d8becf3b6ce89d00d66bbfa6b27bf3cdb8ca8fcd525f0c76d",
11
+ "normalizer_fields": [
12
+ "action",
13
+ "point_cloud"
14
+ ],
15
+ "tensor_bytes": 1020966632,
16
+ "tensor_count": 182
17
+ },
18
+ "run_id": "X1PD1-S42",
19
+ "seed": 42,
20
+ "selection_metric": {
21
+ "mode": "min",
22
+ "name": "pred_cond_Pred_Cond_mean_point-wise_chamfer_distance",
23
+ "scope": "seed-42 validation/top-k selection",
24
+ "value": 6.612324
25
+ },
26
+ "source": {
27
+ "checkpoint_basename": "epoch=40-step=38007-pred_cond_Pred_Cond_mean_point-wise_chamfer_distance=6p6123.ckpt",
28
+ "sha256": "9758331dda8cbc0b1f13365b6c0e4e105dd2436ed7d254884bdc5da00ae11dee"
29
+ },
30
+ "weight_variant": "ema"
31
+ }
cuboids/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0116ddc97f31055d8becf3b6ce89d00d66bbfa6b27bf3cdb8ca8fcd525f0c76d
3
+ size 1020987464
manifest.json ADDED
@@ -0,0 +1,186 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "categories": {
3
+ "containers": {
4
+ "category": "containers",
5
+ "dataset": "containers-v2",
6
+ "epoch": 500,
7
+ "format_version": "3dcov-checkpoint-metadata-v1",
8
+ "global_step": 141783,
9
+ "release": {
10
+ "exact_tensor_roundtrip": true,
11
+ "git_revision": "10403c5d7365d765522f279d892b581d1d4083e5",
12
+ "model_sha256": "19559bfecdba3375d8ebe6fc043a53c10297cedfa226023b484bc2fad8f30102",
13
+ "normalizer_fields": [
14
+ "action",
15
+ "point_cloud"
16
+ ],
17
+ "tensor_bytes": 1020966632,
18
+ "tensor_count": 182
19
+ },
20
+ "run_id": "ODAV4-S42",
21
+ "seed": 42,
22
+ "selection_metric": {
23
+ "mode": "min",
24
+ "name": "pred_cond_Pred_Cond_mean_point-wise_chamfer_distance",
25
+ "scope": "seed-42 validation/top-k selection",
26
+ "value": 347.93262
27
+ },
28
+ "source": {
29
+ "checkpoint_basename": "epoch=500-step=141783-pred_cond_Pred_Cond_mean_point-wise_chamfer_distance=347p9326.ckpt",
30
+ "sha256": "7c66fbf944afa0182e2863b5a163beb09a5696c29333143cf8f25594c5ffbfb8"
31
+ },
32
+ "weight_variant": "ema"
33
+ },
34
+ "cuboids": {
35
+ "category": "cuboids",
36
+ "dataset": "cuboids-v2",
37
+ "epoch": 40,
38
+ "format_version": "3dcov-checkpoint-metadata-v1",
39
+ "global_step": 38007,
40
+ "release": {
41
+ "exact_tensor_roundtrip": true,
42
+ "git_revision": "10403c5d7365d765522f279d892b581d1d4083e5",
43
+ "model_sha256": "0116ddc97f31055d8becf3b6ce89d00d66bbfa6b27bf3cdb8ca8fcd525f0c76d",
44
+ "normalizer_fields": [
45
+ "action",
46
+ "point_cloud"
47
+ ],
48
+ "tensor_bytes": 1020966632,
49
+ "tensor_count": 182
50
+ },
51
+ "run_id": "X1PD1-S42",
52
+ "seed": 42,
53
+ "selection_metric": {
54
+ "mode": "min",
55
+ "name": "pred_cond_Pred_Cond_mean_point-wise_chamfer_distance",
56
+ "scope": "seed-42 validation/top-k selection",
57
+ "value": 6.612324
58
+ },
59
+ "source": {
60
+ "checkpoint_basename": "epoch=40-step=38007-pred_cond_Pred_Cond_mean_point-wise_chamfer_distance=6p6123.ckpt",
61
+ "sha256": "9758331dda8cbc0b1f13365b6c0e4e105dd2436ed7d254884bdc5da00ae11dee"
62
+ },
63
+ "weight_variant": "ema"
64
+ },
65
+ "shelves": {
66
+ "category": "shelves",
67
+ "dataset": "shelves-v2",
68
+ "epoch": 45,
69
+ "format_version": "3dcov-checkpoint-metadata-v1",
70
+ "global_step": 40204,
71
+ "release": {
72
+ "exact_tensor_roundtrip": true,
73
+ "git_revision": "10403c5d7365d765522f279d892b581d1d4083e5",
74
+ "model_sha256": "aa84b6c73a37bb3906df9b60cf69c4942999f0520ef85977e91b1120338e3256",
75
+ "normalizer_fields": [
76
+ "action",
77
+ "point_cloud"
78
+ ],
79
+ "tensor_bytes": 1020966632,
80
+ "tensor_count": 182
81
+ },
82
+ "run_id": "52VCU-S42",
83
+ "seed": 42,
84
+ "selection_metric": {
85
+ "mode": "min",
86
+ "name": "pred_cond_Pred_Cond_mean_point-wise_chamfer_distance",
87
+ "scope": "seed-42 validation/top-k selection",
88
+ "value": 10.069027
89
+ },
90
+ "source": {
91
+ "checkpoint_basename": "epoch=45-step=40204-pred_cond_Pred_Cond_mean_point-wise_chamfer_distance=10p0690.ckpt",
92
+ "sha256": "7a68066d135ab835314779a789305547ecf18de355113b21f917171e4b8358fb"
93
+ },
94
+ "weight_variant": "ema"
95
+ },
96
+ "windows": {
97
+ "category": "windows",
98
+ "dataset": "windows-v2",
99
+ "epoch": 165,
100
+ "format_version": "3dcov-checkpoint-metadata-v1",
101
+ "global_step": 48472,
102
+ "release": {
103
+ "exact_tensor_roundtrip": true,
104
+ "git_revision": "10403c5d7365d765522f279d892b581d1d4083e5",
105
+ "model_sha256": "2ab8dc83ce59b703d37cd269964786c5c61a31c3741e434b90b545d498909d5e",
106
+ "normalizer_fields": [
107
+ "action",
108
+ "point_cloud"
109
+ ],
110
+ "tensor_bytes": 1020966632,
111
+ "tensor_count": 182
112
+ },
113
+ "run_id": "TML4Q-S42",
114
+ "seed": 42,
115
+ "selection_metric": {
116
+ "mode": "min",
117
+ "name": "pred_cond_Pred_Cond_mean_point-wise_chamfer_distance",
118
+ "scope": "seed-42 validation/top-k selection",
119
+ "value": 10.410878
120
+ },
121
+ "source": {
122
+ "checkpoint_basename": "epoch=165-step=48472-pred_cond_Pred_Cond_mean_point-wise_chamfer_distance=10p4109.ckpt",
123
+ "sha256": "bcb0db0b7a5f0c2a4c7bf0472785b5588c4288d627ec212379a37c40b64f14e3"
124
+ },
125
+ "weight_variant": "ema"
126
+ }
127
+ },
128
+ "files": {
129
+ "README.md": {
130
+ "bytes": 3741,
131
+ "sha256": "829a9c0e7c0e93ad101ab7bf8a5d6a587e80d4799384b02cdde62d52fa83b948"
132
+ },
133
+ "containers/config.yaml": {
134
+ "bytes": 1192,
135
+ "sha256": "921479d3af21384ecc0362c9516b019df209f593a1b3ceed985f4d85482e42cc"
136
+ },
137
+ "containers/metrics.json": {
138
+ "bytes": 963,
139
+ "sha256": "7945015f5b38b13551ec707cfa36b39aa1b62233afa9265471c9c864362f04fb"
140
+ },
141
+ "containers/model.safetensors": {
142
+ "bytes": 1020987472,
143
+ "sha256": "19559bfecdba3375d8ebe6fc043a53c10297cedfa226023b484bc2fad8f30102"
144
+ },
145
+ "cuboids/config.yaml": {
146
+ "bytes": 996,
147
+ "sha256": "a405bcbaa8bd80a73353d786e37972ff93f91d5e075aa171d443adec45e3cc4d"
148
+ },
149
+ "cuboids/metrics.json": {
150
+ "bytes": 950,
151
+ "sha256": "7587101b368c5c0de556aa93c1a2e36767428bd073e0f21128733ead1c6496c1"
152
+ },
153
+ "cuboids/model.safetensors": {
154
+ "bytes": 1020987464,
155
+ "sha256": "0116ddc97f31055d8becf3b6ce89d00d66bbfa6b27bf3cdb8ca8fcd525f0c76d"
156
+ },
157
+ "shelves/config.yaml": {
158
+ "bytes": 996,
159
+ "sha256": "a33f9643c53f85d7d6f55a105cf19e6844d194e3bd6535c36701a37c5b002395"
160
+ },
161
+ "shelves/metrics.json": {
162
+ "bytes": 952,
163
+ "sha256": "c2473be1acf7b900db676b87fdbe73688b864523451b1ac246ee5d1a1794dcc6"
164
+ },
165
+ "shelves/model.safetensors": {
166
+ "bytes": 1020987464,
167
+ "sha256": "aa84b6c73a37bb3906df9b60cf69c4942999f0520ef85977e91b1120338e3256"
168
+ },
169
+ "windows/config.yaml": {
170
+ "bytes": 1189,
171
+ "sha256": "b328bc6090444ed4655f64e5d40eade119fb1f9a8f5269195c77484252648c00"
172
+ },
173
+ "windows/metrics.json": {
174
+ "bytes": 954,
175
+ "sha256": "3ddd8f5049bb7312a5416c5684878524450271e59a4b728af227df2d3b775a27"
176
+ },
177
+ "windows/model.safetensors": {
178
+ "bytes": 1020987464,
179
+ "sha256": "2ab8dc83ce59b703d37cd269964786c5c61a31c3741e434b90b545d498909d5e"
180
+ }
181
+ },
182
+ "format_version": "3dcov-model-release-v1",
183
+ "git_revision": "10403c5d7365d765522f279d892b581d1d4083e5",
184
+ "repository": "ChenyuanC/3D-CovDiffusion",
185
+ "weight_variant": "ema"
186
+ }
shelves/config.yaml ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ format_version: 3dcov-inference-config-v1
2
+ task_name: CovDiffusion
3
+ io_type: CovDiffusion
4
+ dataset:
5
+ - shelves-v2
6
+ action_dim: 24
7
+ horizon: 16
8
+ n_action_steps: 100
9
+ n_obs_steps: 1
10
+ encoder_output_dim: 256
11
+ diffusion:
12
+ model_type: dp3
13
+ diffusion_step_embed_dim: 128
14
+ down_dims:
15
+ - 512
16
+ - 1024
17
+ - 2048
18
+ kernel_size: 5
19
+ n_groups: 8
20
+ condition_type: film
21
+ use_down_condition: true
22
+ use_mid_condition: true
23
+ use_up_condition: true
24
+ num_inference_steps: 10
25
+ obs_as_global_cond: true
26
+ model:
27
+ backbone: dp3
28
+ affinetrans: false
29
+ hidden_size:
30
+ - 1024
31
+ - 1024
32
+ pretrained: false
33
+ pretrained_custom: null
34
+ load_strict: false
35
+ noise_scheduler:
36
+ _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
37
+ num_train_timesteps: 100
38
+ beta_start: 0.0001
39
+ beta_end: 0.02
40
+ beta_schedule: squaredcos_cap_v2
41
+ clip_sample: true
42
+ set_alpha_to_one: true
43
+ steps_offset: 0
44
+ prediction_type: sample
45
+ shape_meta:
46
+ obs:
47
+ point_cloud:
48
+ shape:
49
+ - 5120
50
+ - 3
51
+ action:
52
+ shape:
53
+ - 24
shelves/metrics.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "category": "shelves",
3
+ "dataset": "shelves-v2",
4
+ "epoch": 45,
5
+ "format_version": "3dcov-checkpoint-metadata-v1",
6
+ "global_step": 40204,
7
+ "release": {
8
+ "exact_tensor_roundtrip": true,
9
+ "git_revision": "10403c5d7365d765522f279d892b581d1d4083e5",
10
+ "model_sha256": "aa84b6c73a37bb3906df9b60cf69c4942999f0520ef85977e91b1120338e3256",
11
+ "normalizer_fields": [
12
+ "action",
13
+ "point_cloud"
14
+ ],
15
+ "tensor_bytes": 1020966632,
16
+ "tensor_count": 182
17
+ },
18
+ "run_id": "52VCU-S42",
19
+ "seed": 42,
20
+ "selection_metric": {
21
+ "mode": "min",
22
+ "name": "pred_cond_Pred_Cond_mean_point-wise_chamfer_distance",
23
+ "scope": "seed-42 validation/top-k selection",
24
+ "value": 10.069027
25
+ },
26
+ "source": {
27
+ "checkpoint_basename": "epoch=45-step=40204-pred_cond_Pred_Cond_mean_point-wise_chamfer_distance=10p0690.ckpt",
28
+ "sha256": "7a68066d135ab835314779a789305547ecf18de355113b21f917171e4b8358fb"
29
+ },
30
+ "weight_variant": "ema"
31
+ }
shelves/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aa84b6c73a37bb3906df9b60cf69c4942999f0520ef85977e91b1120338e3256
3
+ size 1020987464
windows/config.yaml ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ format_version: 3dcov-inference-config-v1
2
+ task_name: CovDiffusion
3
+ io_type: CovDiffusion
4
+ dataset:
5
+ - windows-v2
6
+ action_dim: 24
7
+ horizon: 16
8
+ n_action_steps: 100
9
+ n_obs_steps: 1
10
+ encoder_output_dim: 256
11
+ shape_meta:
12
+ obs:
13
+ point_cloud:
14
+ shape:
15
+ - 5120
16
+ - 3
17
+ low_dim:
18
+ shape:
19
+ - 24
20
+ action:
21
+ shape:
22
+ - 24
23
+ diffusion:
24
+ model_type: dp3
25
+ diffusion_step_embed_dim: 128
26
+ down_dims:
27
+ - 512
28
+ - 1024
29
+ - 2048
30
+ kernel_size: 5
31
+ n_groups: 8
32
+ condition_type: film
33
+ use_down_condition: true
34
+ use_mid_condition: true
35
+ use_up_condition: true
36
+ num_inference_steps: 10
37
+ obs_as_global_cond: true
38
+ model:
39
+ backbone: dp3
40
+ affinetrans: false
41
+ hidden_size:
42
+ - 1024
43
+ - 1024
44
+ pretrained: false
45
+ pretrained_custom: null
46
+ load_strict: false
47
+ noise_scheduler:
48
+ _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
49
+ num_train_timesteps: 100
50
+ beta_start: 0.0001
51
+ beta_end: 0.02
52
+ beta_schedule: squaredcos_cap_v2
53
+ clip_sample: true
54
+ set_alpha_to_one: true
55
+ steps_offset: 0
56
+ prediction_type: sample
57
+ partial_observation_enabled: false
58
+ partial_observation_method: fixed_camera
59
+ partial_observation_ratio: 0.3
60
+ ablation_prev_traj: normal
61
+ random_traj_seed: 123
windows/metrics.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "category": "windows",
3
+ "dataset": "windows-v2",
4
+ "epoch": 165,
5
+ "format_version": "3dcov-checkpoint-metadata-v1",
6
+ "global_step": 48472,
7
+ "release": {
8
+ "exact_tensor_roundtrip": true,
9
+ "git_revision": "10403c5d7365d765522f279d892b581d1d4083e5",
10
+ "model_sha256": "2ab8dc83ce59b703d37cd269964786c5c61a31c3741e434b90b545d498909d5e",
11
+ "normalizer_fields": [
12
+ "action",
13
+ "point_cloud"
14
+ ],
15
+ "tensor_bytes": 1020966632,
16
+ "tensor_count": 182
17
+ },
18
+ "run_id": "TML4Q-S42",
19
+ "seed": 42,
20
+ "selection_metric": {
21
+ "mode": "min",
22
+ "name": "pred_cond_Pred_Cond_mean_point-wise_chamfer_distance",
23
+ "scope": "seed-42 validation/top-k selection",
24
+ "value": 10.410878
25
+ },
26
+ "source": {
27
+ "checkpoint_basename": "epoch=165-step=48472-pred_cond_Pred_Cond_mean_point-wise_chamfer_distance=10p4109.ckpt",
28
+ "sha256": "bcb0db0b7a5f0c2a4c7bf0472785b5588c4288d627ec212379a37c40b64f14e3"
29
+ },
30
+ "weight_variant": "ema"
31
+ }
windows/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2ab8dc83ce59b703d37cd269964786c5c61a31c3741e434b90b545d498909d5e
3
+ size 1020987464