Add files using upload-large-folder tool

config.yaml

@@ -0,0 +1,587 @@
+root_dir: /mnt/kostas-graid/datasets/extrinsic_contact_data
+replay_buffer_size: 150000
+replay_buffer_num_workers: 2
+nstep: 3
+batch_size: 128
+seed: 0
+dataset_shuffle_seed: ${seed}
+valid_dataset_shuffle_seed: 0
+train_dataset_shuffle_seed: 2
+device: cuda
+save_video: true
+save_train_video: true
+use_tb: true
+use_wandb: true
+wandb_run_id: '36309_0'
+wandb_notes: 36309_0_restarted_5
+eval: false
+eval_on_sim: true
+simulation: true
+true_action_history: false
+train_pad_after: 4
+process_contact_features: ${eval}
+obs_type: pixels
+use_color: true
+use_depth: true
+use_masks: false
+mask_list:
+- EE_obj_mask
+mask_representation: channels
+segmentation_model_name: gt_segmentation
+crop_hw:
+- 144
+- 144
+crop_down_offset: 48
+color_crop_type: null
+depth_crop_type: null
+segmask_crop_type: null
+add_crop_binary_mask: false
+add_coord_conv_map: false
+use_context_color: false
+use_context_depth: false
+use_context_segmask: false
+context_color_crop_type: null
+context_depth_crop_type: null
+context_segmask_crop_type: null
+context_add_crop_binary_mask: false
+context_add_coord_conv_map: false
+use_contact_map: true
+use_sdf_maps: false
+use_normals_maps: false
+which_objects: both
+max_contact_prob: 1.0
+max_depth: 2.0
+grasped_dtc_max_value: 0.2
+env_dtc_max_value: 0.4
+grasped_normals_mask_max_dtc_value: 0.2
+env_normals_mask_max_dtc_value: 0.4
+clamp_dtc: true
+dtc_adaptive_normalization: false
+mask_normals_within_sdf: true
+adaptive_normals_mask: true
+learnable_contact_preprocess_params: true
+contact_model_name: gt_contact
+contact_estimation_model_ckpt_path: ~/fish_leon/contact_estimation/artifacts/175604_2/checkpoints/epoch=09-val_loss=0.00.ckpt
+encoder_type: small
+debug_timestamps: false
+open_loop: false
+action_trajectories: true
+stop_after_action: false
+interpolation_frequency: 25
+policy_frequency: 5
+wait_for_new_camera_frames: true
+baseline: false
+train_demo_idxs_list_or_num: -1
+log_train_every_steps: 25
+name_of_expert_demo: 413_sim_demos_left_of_4th_book_20hz_act
+expert_dataset_dirpath: ${root_dir}/FISH/expert_demos/${suite.name}/${task_name}/${name_of_expert_demo}
+store_dataset_in_memory: false
+expert_dataset: ${expert_dataset_dirpath}/demos.zarr
+action_key: action
+semantic_demo_grouping_name: semantic_demo_grouping.yaml
+semantic_demo_grouping: ${expert_dataset_dirpath}/${semantic_demo_grouping_name}
+include_groups_list: all
+expert_dataset_config: ${expert_dataset_dirpath}/demo_config.yaml
+name_of_valid_demo: 413_sim_demos_left_of_4th_book_20hz_act
+train_num_groups: 280
+valid_dataset_dir: ${root_dir}/FISH/expert_demos/${suite.name}/${task_name}/${name_of_valid_demo}/demos.zarr
+valid_demo_idxs_list_or_num: null
+val_num_groups: 30
+load_bc: ${agent.load_checkpoint}
+checkpoint_epoch_list:
+- 99
+- 199
+- 299
+- 399
+- 499
+- 599
+- 699
+- 799
+- 899
+- 999
+- 1249
+- 1499
+- 1749
+- 1999
+- 2999
+- 3999
+- 4999
+- 5999
+- 6999
+- 7999
+- 8999
+- 9999
+snapshot_root_dir: /mnt/grasp_high_usage/leonmkim/contact_estimation/FISH
+save_snapshot: true
+save_last_snapshot: true
+save_snapshot_when_done: true
+top_k_checkpoints: 5
+save_snapshot_link_to_weights_dir: deprecated
+restore_missing_checkpoint: null
+bc_regularize: false
+bc_weight_type: qfilter
+username: leonmkim
+hostname: grasp-login1
+local_snapshot_root_dir: /mnt/bighdd/fish_contact_backup
+local_snapshot_dir: ${local_snapshot_root_dir}/exp_local/${suite.name}_${obs_type}/${task_name}
+experiment_dir: ./exp_local/${suite.name}_${obs_type}/${task_name}/${wandb_run_id}
+agent:
+  _target_: agent.diffusion_policy.DiffusionPolicyAgent
+  name: diffusion_policy
+  load_checkpoint: ${eval}
+  device: ${device}
+  n_obs_steps: ${.config.policy_cfg.n_obs_steps}
+  suite_name: ${suite.name}
+  obs_type: ${obs_type}
+  enable_arm: ${eval}
+  enable_camera: ${eval}
+  use_tb: ${use_tb}
+  desired_image_shape:
+  - 13
+  - 180
+  - 240
+  orig_cam_shape:
+  - 3
+  - 240
+  - 320
+  actions_prenormalized: ${simulation}
+  config:
+    _target_: agent.diffusion_policy.DiffusionPolicyAgentConfig
+    compile: false
+    device: ${device}
+    cam_resize_shape: ${agent.desired_image_shape}
+    orig_cam_shape: ${agent.orig_cam_shape}
+    policy_frequency: ${policy_frequency}
+    interpolation_frequency: ${interpolation_frequency}
+    policy_cfg:
+      _target_: lerobot.common.policies.diffusion.configuration_diffusion.DiffusionConfig
+      n_obs_steps: 1
+      horizon: 28
+      n_action_steps: ${agent.config.policy_cfg.horizon}
+      output_shapes:
+        action:
+        - 7
+      input_normalization_modes:
+        observation.image: mean_std
+        observation.state: min_max
+        observation.action_history: min_max
+      output_normalization_modes:
+        action: min_max
+      vision_backbone: resnet18
+      crop_distractors_tlhw: null
+      pretrained_backbone_weights: null
+      transforms:
+      - _target_: torchaug.transforms.RandomAffine
+        degrees:
+        - -5
+        - 5
+        translate:
+        - 0.05
+        - 0.05
+        batch_transform: true
+        num_chunks: -1
+        batch_inplace: true
+      - _target_: torchaug.transforms.RandomColorJitter
+        brightness: 0.3
+        contrast: 0.4
+        saturation: 0.5
+        hue: 0.08
+        batch_transform: true
+        num_chunks: -1
+        batch_inplace: true
+      use_group_norm: true
+      spatial_softmax_num_keypoints: 32
+      action_history_encoder_config:
+        _target_: lerobot.common.policies.diffusion.configuration_diffusion.Unet1dEncoderConfig
+        in_channels: 7
+        out_channels: 32
+        history_length: 4
+        kernel_size: ${agent.config.policy_cfg.kernel_size}
+        downsample_kernel_size: 3
+        downsample_stride: 2
+        downsample_padding: 1
+      down_dims:
+      - 64
+      - 128
+      - 256
+      kernel_size: 5
+      n_groups: 8
+      diffusion_step_embed_dim: 128
+      use_film_scale_modulation: true
+      noise_scheduler_type: DDIM
+      beta_schedule: squaredcos_cap_v2
+      beta_start: 0.0001
+      beta_end: 0.02
+      prediction_type: epsilon
+      clip_sample: true
+      clip_sample_range: 1.0
+      num_train_timesteps: 50
+      num_inference_steps: 10
+      do_mask_loss_for_padding: false
+      input_shapes:
+        observation.image:
+        - 13
+        - 180
+        - 240
+        context_observation.image:
+        - 13
+        - 180
+        - 240
+        observation.state:
+        - 8
+        observation.action_history:
+        - 7
+    train_cfg:
+      _target_: utils.TrainConfig
+      lr: 0.0001
+      lr_scheduler: cosine
+      lr_warmup_steps: 500
+      adam_betas:
+      - 0.95
+      - 0.999
+      adam_eps: 1.0e-08
+      adam_weight_decay: 1.0e-06
+      grad_clip_norm: 10
+      offline_steps: ${num_train_frames_diffusion}
+      use_amp: true
+    observation_cfg:
+      _target_: agent.encoder.VisualFeatureSet
+      use_depth: ${use_depth}
+      use_color: ${use_color}
+      mask_input_dict:
+        _target_: agent.encoder.MaskInputDict
+        enable: ${use_masks}
+        representation: ${mask_representation}
+        mask_list: ${mask_list}
+        segmentation_model_name: ${segmentation_model_name}
+      crop_input_config:
+        _target_: agent.encoder.CropInputConfig
+        color_crop_type: ${color_crop_type}
+        depth_crop_type: ${depth_crop_type}
+        segmask_crop_type: ${segmask_crop_type}
+        crop_hw: ${crop_hw}
+        crop_down_offset: ${crop_down_offset}
+        add_crop_binary_mask: ${add_crop_binary_mask}
+        add_coord_conv_map: ${add_coord_conv_map}
+      context_input_config:
+        _target_: agent.encoder.ContextInputConfig
+        use_color: ${use_context_color}
+        use_depth: ${use_context_depth}
+        mask_input_dict:
+          _target_: agent.encoder.MaskInputDict
+          enable: ${use_context_segmask}
+          representation: ${mask_representation}
+          mask_list: ${mask_list}
+          segmentation_model_name: ${segmentation_model_name}
+        crop_input_config:
+          _target_: agent.encoder.CropInputConfig
+          color_crop_type: ${context_color_crop_type}
+          depth_crop_type: ${context_depth_crop_type}
+          segmask_crop_type: ${context_segmask_crop_type}
+          crop_hw: ${crop_hw}
+          crop_down_offset: ${crop_down_offset}
+          add_crop_binary_mask: ${context_add_crop_binary_mask}
+          add_coord_conv_map: ${context_add_coord_conv_map}
+      mask_soft_approx_scheduler_config:
+        _target_: agent.encoder.MaskSoftApproxSchedulerConfig
+        num_steps: 40000
+        initial_value: 10.0
+        final_value: 1000.0
+        interpolation_scheme: cosine
+      use_contact_map: ${use_contact_map}
+      use_sdf_maps: ${use_sdf_maps}
+      use_normals_maps: ${use_normals_maps}
+      which_objects: ${which_objects}
+      grasped_dtc_max_value: ${grasped_dtc_max_value}
+      env_dtc_max_value: ${env_dtc_max_value}
+      grasped_normals_mask_max_dtc_value: ${grasped_normals_mask_max_dtc_value}
+      env_normals_mask_max_dtc_value: ${env_normals_mask_max_dtc_value}
+      clamp_dtc: ${clamp_dtc}
+      max_contact_prob: ${max_contact_prob}
+      mask_normals_within_sdf: ${mask_normals_within_sdf}
+      dtc_adaptive_normalization: ${dtc_adaptive_normalization}
+      adaptive_normals_mask: ${adaptive_normals_mask}
+      max_depth: ${max_depth}
+      image_shape: ${agent.desired_image_shape}
+      learnable_contact_preprocess_params: ${learnable_contact_preprocess_params}
+      learning_rate: 0.0001
+      weight_decay: 0.0
+      contact_model_name: ${contact_model_name}
+      zero_centered: false
+      crop_distractors_tlhw: ${agent.config.policy_cfg.crop_distractors_tlhw}
+suite:
+  suite: frankagym
+  name: frankagym
+  frame_stack: ${agent.n_obs_steps}
+  action_repeat: 1
+  discount: 0.99
+  hidden_dim: 1024
+  num_train_frames: 1000000
+  num_seed_frames: 0
+  num_train_epochs: 500
+  validate_every_epochs: 25
+  eval_on_sim_every_epochs: 25
+  validate_diffusion_on_action_loss_every_epochs: 25
+  eval_on_sim_every_epochs_diffusion: 500
+  train_eval_diffusion_on_action_loss_every_epochs: 25
+  check_topk_every_epochs: 25
+  save_snapshot_every_epochs: 50
+  num_seen_eval_envs: 25
+  num_unseen_eval_envs: 25
+  evals_per_env: 1
+  eval_every_frames: 2000
+  num_eval_episodes: 5
+  save_snapshot: true
+  wait_for_user_to_start_episode: true
+  task_make_fn:
+    _target_: suite.frankagym.make
+    name: ${task_name}
+    height: 240
+    width: 320
+    frame_stack: ${suite.frame_stack}
+    action_repeat: ${suite.action_repeat}
+    seed: ${seed}
+    enable_arm: ${agent.enable_arm}
+    enable_gripper: ${enable_gripper}
+    start_with_gripper_open: ${start_with_gripper_open}
+    enable_camera: ${agent.enable_camera}
+    path_to_depth_extrinsics: ${path_to_depth_extrinsics}
+    contact_estimation_model_ckpt_path: ${contact_estimation_model_ckpt_path}
+    x_limit: ${x_limit}
+    y_limit: ${y_limit}
+    z_limit: ${z_limit}
+    device: ${device}
+    interpolation_frequency: ${interpolation_frequency}
+    policy_frequency: ${policy_frequency}
+    debug_timestamps: ${debug_timestamps}
+    stop_after_action: ${stop_after_action}
+    open_loop: ${open_loop}
+    wait_for_new_camera_frames: ${wait_for_new_camera_frames}
+    action_key: ${action_key}
+    action_trajectory_horizon: ${agent.config.policy_cfg.horizon}
+    action_trajectories: ${action_trajectories}
+    path_to_zarr_dataset: ${expert_dataset}
+    agent_policy_cfg:
+      _target_: agent.diffusion_policy.DiffusionPolicyAgentConfig
+      compile: false
+      device: ${device}
+      cam_resize_shape: ${agent.desired_image_shape}
+      orig_cam_shape: ${agent.orig_cam_shape}
+      policy_frequency: ${policy_frequency}
+      interpolation_frequency: ${interpolation_frequency}
+      policy_cfg:
+        _target_: lerobot.common.policies.diffusion.configuration_diffusion.DiffusionConfig
+        n_obs_steps: 1
+        horizon: 28
+        n_action_steps: ${agent.config.policy_cfg.horizon}
+        output_shapes:
+          action:
+          - 7
+        input_normalization_modes:
+          observation.image: mean_std
+          observation.state: min_max
+          observation.action_history: min_max
+        output_normalization_modes:
+          action: min_max
+        vision_backbone: resnet18
+        crop_distractors_tlhw: null
+        pretrained_backbone_weights: null
+        transforms:
+        - _target_: torchaug.transforms.RandomAffine
+          degrees:
+          - -5
+          - 5
+          translate:
+          - 0.05
+          - 0.05
+          batch_transform: true
+          num_chunks: -1
+          batch_inplace: true
+        - _target_: torchaug.transforms.RandomColorJitter
+          brightness: 0.3
+          contrast: 0.4
+          saturation: 0.5
+          hue: 0.08
+          batch_transform: true
+          num_chunks: -1
+          batch_inplace: true
+        use_group_norm: true
+        spatial_softmax_num_keypoints: 32
+        action_history_encoder_config:
+          _target_: lerobot.common.policies.diffusion.configuration_diffusion.Unet1dEncoderConfig
+          in_channels: 7
+          out_channels: 32
+          history_length: 4
+          kernel_size: ${agent.config.policy_cfg.kernel_size}
+          downsample_kernel_size: 3
+          downsample_stride: 2
+          downsample_padding: 1
+        down_dims:
+        - 64
+        - 128
+        - 256
+        kernel_size: 5
+        n_groups: 8
+        diffusion_step_embed_dim: 128
+        use_film_scale_modulation: true
+        noise_scheduler_type: DDIM
+        beta_schedule: squaredcos_cap_v2
+        beta_start: 0.0001
+        beta_end: 0.02
+        prediction_type: epsilon
+        clip_sample: true
+        clip_sample_range: 1.0
+        num_train_timesteps: 50
+        num_inference_steps: 10
+        do_mask_loss_for_padding: false
+        input_shapes:
+          observation.image:
+          - 13
+          - 180
+          - 240
+          context_observation.image:
+          - 13
+          - 180
+          - 240
+          observation.state:
+          - 8
+          observation.action_history:
+          - 7
+      train_cfg:
+        _target_: utils.TrainConfig
+        lr: 0.0001
+        lr_scheduler: cosine
+        lr_warmup_steps: 500
+        adam_betas:
+        - 0.95
+        - 0.999
+        adam_eps: 1.0e-08
+        adam_weight_decay: 1.0e-06
+        grad_clip_norm: 10
+        offline_steps: ${num_train_frames_diffusion}
+        use_amp: true
+      observation_cfg:
+        _target_: agent.encoder.VisualFeatureSet
+        use_depth: ${use_depth}
+        use_color: ${use_color}
+        mask_input_dict:
+          _target_: agent.encoder.MaskInputDict
+          enable: ${use_masks}
+          representation: ${mask_representation}
+          mask_list: ${mask_list}
+          segmentation_model_name: ${segmentation_model_name}
+        crop_input_config:
+          _target_: agent.encoder.CropInputConfig
+          color_crop_type: ${color_crop_type}
+          depth_crop_type: ${depth_crop_type}
+          segmask_crop_type: ${segmask_crop_type}
+          crop_hw: ${crop_hw}
+          crop_down_offset: ${crop_down_offset}
+          add_crop_binary_mask: ${add_crop_binary_mask}
+          add_coord_conv_map: ${add_coord_conv_map}
+        context_input_config:
+          _target_: agent.encoder.ContextInputConfig
+          use_color: ${use_context_color}
+          use_depth: ${use_context_depth}
+          mask_input_dict:
+            _target_: agent.encoder.MaskInputDict
+            enable: ${use_context_segmask}
+            representation: ${mask_representation}
+            mask_list: ${mask_list}
+            segmentation_model_name: ${segmentation_model_name}
+          crop_input_config:
+            _target_: agent.encoder.CropInputConfig
+            color_crop_type: ${context_color_crop_type}
+            depth_crop_type: ${context_depth_crop_type}
+            segmask_crop_type: ${context_segmask_crop_type}
+            crop_hw: ${crop_hw}
+            crop_down_offset: ${crop_down_offset}
+            add_crop_binary_mask: ${context_add_crop_binary_mask}
+            add_coord_conv_map: ${context_add_coord_conv_map}
+        mask_soft_approx_scheduler_config:
+          _target_: agent.encoder.MaskSoftApproxSchedulerConfig
+          num_steps: 40000
+          initial_value: 10.0
+          final_value: 1000.0
+          interpolation_scheme: cosine
+        use_contact_map: ${use_contact_map}
+        use_sdf_maps: ${use_sdf_maps}
+        use_normals_maps: ${use_normals_maps}
+        which_objects: ${which_objects}
+        grasped_dtc_max_value: ${grasped_dtc_max_value}
+        env_dtc_max_value: ${env_dtc_max_value}
+        grasped_normals_mask_max_dtc_value: ${grasped_normals_mask_max_dtc_value}
+        env_normals_mask_max_dtc_value: ${env_normals_mask_max_dtc_value}
+        clamp_dtc: ${clamp_dtc}
+        max_contact_prob: ${max_contact_prob}
+        mask_normals_within_sdf: ${mask_normals_within_sdf}
+        dtc_adaptive_normalization: ${dtc_adaptive_normalization}
+        adaptive_normals_mask: ${adaptive_normals_mask}
+        max_depth: ${max_depth}
+        image_shape: ${agent.desired_image_shape}
+        learnable_contact_preprocess_params: ${learnable_contact_preprocess_params}
+        learning_rate: 0.0001
+        weight_decay: 0.0
+        contact_model_name: ${contact_model_name}
+        zero_centered: false
+        crop_distractors_tlhw: ${agent.config.policy_cfg.crop_distractors_tlhw}
+    true_action_history: ${true_action_history}
+num_train_frames_bc: 50000
+num_train_frames_drq: 1100000
+stddev_schedule_drq: linear(1.0,0.1,100000)
+task_name: FrankaInsertion-v1
+num_train_frames_vinn: 25000
+num_train_frames_diffusion: 1000000
+num_train_epochs_bc: 5000
+num_train_epochs_diffusion: 500
+validate_every_epochs_bc: 5
+validate_every_epochs_diffusion: 25
+validate_diffusion_on_action_loss_every_epochs: 25
+train_eval_diffusion_on_action_loss_every_epochs: 25
+eval_on_sim_every_epochs_diffusion: 25
+check_topk_every_epochs: 5
+check_topk_every_epochs_diffusion: 25
+save_snapshot_every_epochs_diffusion: 50
+x_limit:
+- 0.2
+- 0.7
+y_limit:
+- -0.4
+- 0.4
+z_limit:
+- -0.05
+- 0.55
+home_displacement:
+- 0.55
+- 0.0
+- 0.55
+- 180.0
+- 0.0
+- 0.0
+enable_gripper: true
+start_with_gripper_open: true
+offset_mask:
+- 1
+- 1
+- 1
+- 1
+- 1
+- 1
+path_to_depth_extrinsics: ~/fish_leon/FISH/cfgs/camera_poses/camera_poses_L515/20240904-122305/color_tf_world.npy
+test:
+  username: leonmkim
+  hostname: grasp-login1
+  name_of_expert_demo: 112_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act
+  root_dir: ~/fish_leon
+  expert_dataset_dirpath: ${test.root_dir}/FISH/expert_demos/${suite.name}/${task_name}/${test.name_of_expert_demo}
+  expert_dataset: ${test.expert_dataset_dirpath}/demos.zarr
+  semantic_demo_grouping_name: semantic_demo_grouping.yaml
+  semantic_demo_grouping: ${test.expert_dataset_dirpath}/${test.semantic_demo_grouping_name}
+  include_groups_list: all
+  batch_size: 128
+  store_dataset_in_memory: false
+  use_tb: true
+  local_snapshot_root_dir: /mnt/bighdd/fish_contact_backup
+  local_snapshot_dir: ${test.local_snapshot_root_dir}/exp_local/${suite.name}_${obs_type}/${task_name}
+  resume_wandb_run: false
+feature_type: 180x240_1_RGB_D_2.0_cntct_1.0_gt_contact_acthist_hst4_out32_dwnkrnl3_dwnstrd2_dwnpd1

snapshot_300.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d330ee596568eb59f2d3e4982e81ff9d3e93c93dfc94f6c45dc2850f8734a58
+size 192469798