current_works_v2: layout-only guidance weight sweep (20 configs, 370 test scenes)

current_works_v2/sweep_layout_only/README.md

@@ -0,0 +1,9 @@
+# Guidance weight sweep (size-freeze fix, layout-only, 370 test scenes)
+
+All runs: branch `fix/guidance-freeze-sizes`, server `~/LHAN/SteerScene/echoscene_sizefix`, base config
+`config/work41_sizefix.yaml` + overrides (`sweep_round1.tsv` on the server), layout only (`--gen_shape False`),
+same checkpoint and FRONT data. Score with `./score_sweep.sh sweep_round1.tsv` → `table.md`.
+
+Selection rule (fixed before looking at results): among configs whose SG Total after PP is not below the
+baseline's, pick the one with the lowest raw ColObj, then lowest raw OOB; PP displacement breaks ties.
+Amended 2026-10-20 (user): a config qualifies if its SG Total beats the baseline either raw or after PP.

current_works_v2/sweep_layout_only/configs/baseline.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: false
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 10.0
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 10.0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_c_col100.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 100
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: false
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_c_col30.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 30
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: false
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_col.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 10.0
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: false
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_col_rel.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 10.0
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_col_room.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 10.0
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 10.0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: false
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_col_room_rel.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 10.0
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 10.0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_col_room_rel_walk.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 10.0
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 10.0
+      walkable:
+        enabled: true
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_cr_col100.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 100
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_cr_col30.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 30
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_cr_col30_rel30.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 30
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 30
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_crr_col100.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 100
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 10.0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_crr_col100_room1.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 100
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 1
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_crr_col30.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 30
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 10.0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_crr_col30_room1.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 30
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 1
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_crr_col30_room3.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 30
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 3
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_crr_rel30.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 10.0
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 10.0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 30
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_crr_start50.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.5
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 20.0
+        weight: 10.0
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 10.0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_crr_str50.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: true
+        strength: 50
+        weight: 10.0
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 10.0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0

current_works_v2/sweep_layout_only/configs/lo_room_rel.yaml

@@ -0,0 +1,134 @@
+hyper:
+  batch_size: 16
+  gpu_ids: 0
+  logs_dir: ../full_mp/diff_crossattn
+  results_dir: ../full_mp/diff_crossattn
+  name: ./
+  isTrain: true
+  device: cuda
+  distributed: 0
+  lr_init: 0.0001
+  lr_step:
+  - 35000
+  - 70000
+  - 140000
+  lr_evo:
+  - 5.0e-05
+  - 1.0e-05
+  - 5.0e-06
+dataset:
+  res: 64
+  trunc_thres: 0.2
+  ratio: 1
+layout_branch:
+  model: diffusion_scene_layout_ddpm
+  angle_dim: 2
+  denoiser: unet1d
+  relation_condition: true
+  denoiser_kwargs:
+    dims: 1
+    in_channels: 8
+    out_channels: 8
+    model_channels: 512
+    channel_mult:
+    - 1
+    - 1
+    - 1
+    - 1
+    num_res_blocks: 2
+    attention_resolutions:
+    - 4
+    - 2
+    num_heads: 8
+    use_spatial_transformer: true
+    transformer_depth: 1
+    conditioning_key: crossattn
+    concat_dim: 1280
+    crossattn_dim: 1280
+    use_checkpoint: true
+    enable_t_emb: true
+  diffusion_kwargs:
+    schedule_type: linear
+    beta_start: 0.0001
+    beta_end: 0.02
+    time_num: 1000
+    model_mean_type: eps
+    model_var_type: fixedsmall
+    loss_separate: true
+    loss_iou: false
+    iou_type: obb
+    train_stats_file: null
+  inference_guidance:
+    enabled: true
+    guide_sizes: false
+    interval: 1
+    start_ratio: 0.9
+    grad_clip: 10.1
+    clip_to_bounds: true
+    constraints:
+      collision:
+        enabled: false
+        strength: 20.0
+        weight: 10.0
+        iou_threshold: 0.02
+        penetration_threshold: 0.0
+        iou_weight: 0.5
+        penetration_weight: 2.0
+        metric: aabb
+        ignore_labels:
+        - scene
+        - floor
+      room_outer:
+        weight: 10.0
+      walkable:
+        enabled: false
+        weight: 1.0
+        components:
+          center_penalty:
+            enabled: true
+            weight: 1
+            sigma: 0.5
+          pathfinding:
+            enabled: false
+            weight: 0.1
+            robot_width_real: 0.5
+            robot_hight_real: 1.5
+          edge_gaussian:
+            enabled: true
+            weight: 1
+            robot_width_real: 0.35
+            robot_hight_real: 1.5
+            sigma_scale: 0.5
+            heatmap_weight: 0.8
+            repulsion_weight: 0.2
+      relational:
+        enabled: true
+        weight: 10.0
+        start_ratio: 0.5
+        interval: 1
+        margin: 0.05
+        close_threshold: 0.45
+        stand_threshold: 0.04
+shape_branch:
+  model: sdfusion-txt2shape_mp
+  sampling: greedy
+  ckpt: null
+  df_cfg: ../config/sdfusion-txt2shape_mp.yaml
+  ddim_steps: 100
+  ddim_eta: 0.0
+  uc_scale: 3.0
+  vq_model: vqvae
+  vq_cfg: ../config/vqvae_snet.yaml
+  vq_dset: None
+  vq_cat: None
+  vq_ckpt: ./checkpoint/vqvae_threedfront_best.pth
+misc:
+  debug: 0
+  seed: 111
+  backend: gloo
+  local_rank: 0
+training:
+  lr: 1.0e-05
+  lr_policy: lambda
+  lr_decay_iters: 50
+  lambda_L1: 10.0