confs_new/dyn_grab_pointset_points_dyn_retar_pts.conf

general {


    base_exp_dir = exp/CASE_NAME/wmask
    
    tag = "train_retargeted_shadow_hand_seq_102_mano_pointset_acts_"
    tag = "train_retargeted_shadow_hand_seq_102_mano_pointset_acts_retar_to_shadow_pointset_"

    recording = [
        ./,
        ./models
    ]
}

dataset {
    data_dir = public_data/CASE_NAME/
    render_cameras_name = cameras_sphere.npz
    object_cameras_name = cameras_sphere.npz
    obj_idx = 102
}

train {
    learning_rate = 5e-4
    learning_rate_alpha = 0.05
    end_iter = 300000

    batch_size = 1024 #  64
    validate_resolution_level = 4
    warm_up_end = 5000
    anneal_end = 0
    use_white_bkgd = False

    # save_freq = 10000
    save_freq = 10000
    val_freq = 20 #  2500
    val_mesh_freq = 20 #  5000
    report_freq = 10
    ### igr weight ###
    igr_weight = 0.1
    mask_weight = 0.1
}
 
model {

    optimize_dyn_actions = True


    optimize_robot = True

    use_penalty_based_friction = True

    use_split_params = False

    use_sqr_spring_stiffness = True

    use_pre_proj_frictions = True



    use_sqrt_dist = True
    contact_maintaining_dist_thres = 0.2

    robot_actions_diff_coef = 0.001


    use_sdf_as_contact_dist = True


    # 
    use_contact_dist_as_sdf = False

    use_glb_proj_delta = True



# penetration_proj_k_to_robot = 30
    penetrating_depth_penalty = 1.0
    train_states = True



    minn_dist_threshold = 0.000
    obj_mass = 30.0


    use_LBFGS = True
    use_LBFGS = False

    use_mano_hand_for_test = False # use the dynamic mano model here #

    extract_delta_mesh = False
    freeze_weights = True
    gt_act_xs_def = False
    use_bending_network =  True
    ### for ts = 3 ###
    # use_delta_bending = False
    ### for ts = 3 ###
   



    sim_model_path = "rsc/shadow_hand_description/shadowhand_new.urdf"
    mano_sim_model_path = "rsc/mano/mano_mean_wcollision_scaled_scaled_0_9507_nroot.urdf"

    obj_sdf_fn = "data/grab/102/102_obj.npy" 
    kinematic_mano_gt_sv_fn = "data/grab/102/102_sv_dict.npy"
    scaled_obj_mesh_fn = "data/grab/102/102_obj.obj"

    bending_net_type = "active_force_field_v18"    
    sim_num_steps = 1000000
    n_timesteps = 60
    optim_sim_model_params_from_mano = False
    penetration_determining = "sdf_of_canon"
    train_with_forces_to_active = False
    loss_scale_coef = 1000.0
    use_same_contact_spring_k = False
    use_optimizable_params = True # 
    train_residual_friction = True
    mano_mult_const_after_cent = 1.0
    optimize_glb_transformations = True
    no_friction_constraint = False
    optimize_active_object = True
    loss_tangential_diff_coef = 0
    optimize_with_intermediates = True
    using_delta_glb_trans = False
    train_multi_seqs = False
    use_split_network = True
    use_delta_bending = True









    ###### threshold, ks settings 1, optimize acts ######
    # drive_pointset = "actions"
    # fix_obj = True
    # optimize_rules = False
    # train_pointset_acts_via_deltas = True
    # load_optimized_init_actions = "/data/xueyi/NeuS/exp/hand_test_routine_2_light_color_wtime_active_passive/wmask_reverse_value_totviews_tag_train_dyn_mano_hand_seq_102_mouse_optdynactions_points_optrobo_offsetdriven_optrules_multk100_wfixobj_optdelta_radius0d4_/checkpoints/ckpt_002000.pth"
    # load_optimized_init_actions = "/data/xueyi/NeuS/exp/hand_test_routine_2_light_color_wtime_active_passive/wmask_reverse_value_totviews_tag_train_dyn_mano_hand_seq_102_mouse_optdynactions_points_optrobo_offsetdriven_optrules_multk100_wfixobj_optdelta_radius0d2_/checkpoints/ckpt_008000.pth"
    ###### threshold, ks settings 1, optimize acts ######


    ##### contact spring model settings ####
    minn_dist_threshold_robot_to_obj = 0.1
    penetration_proj_k_to_robot_friction = 10000000.0
    penetration_proj_k_to_robot = 4000000.0
    ##### contact spring model settings ####
 

    # ###### Stage 1: threshold, ks settings 1, optimize offsets ######
    # drive_pointset = "states"
    # fix_obj = True
    # optimize_rules = False
    # train_pointset_acts_via_deltas = False
    # load_optimized_init_actions = "ckpts/grab/102/dyn_mano_arti.pth"
    # ###### Stage 1: threshold, ks settings 1, optimize offsets ######


    # ###### Stage 2: threshold, ks settings 1, optimize acts ######
    # drive_pointset = "actions"
    # fix_obj = True
    # optimize_rules = False
    # train_pointset_acts_via_deltas = True
    # load_optimized_init_actions = "ckpts/grab/102/dyn_mano_pointset_states.pt"
    # ###### Stage 2: threshold, ks settings 1, optimize acts ######


    # ###### Stage 3: threshold, ks settings 1, optimize params from acts ######
    # drive_pointset = "actions"
    # fix_obj = False
    # optimize_rules = True
    # train_pointset_acts_via_deltas = True
    # load_optimized_init_actions = "ckpts/grab/102/dyn_mano_pointset_acts.pt"
    # ##### model parameters optimized from the MANO hand trajectory #####
    # ckpt_fn = "ckpts/grab/102/dyn_mano_opts.pt"
    # ###### Stage 3: threshold, ks settings 1, optimize params from acts ######


    # ###### Stage 4: threshold, ks settings 1, optimize acts from optimized params ######
    # drive_pointset = "actions"
    # fix_obj = False
    # optimize_rules = False
    # train_pointset_acts_via_deltas = True ## pointset acts via deltas ###
    # ##### model parameters optimized from the MANO hand expanded set trajectory #####
    # ckpt_fn = "ckpts/grab/102/dyn_mano_pointset_optimized_acts_optimized_ps.pth"
    # load_optimized_init_actions = "ckpts/grab/102/dyn_mano_pointset_optimized_acts.pth"
    # ###### Stage 4: threshold, ks settings 1, optimize acts from optimized params ######



    ###### Retargeting Stage 1 ######
    drive_pointset = "actions"
    fix_obj = False
    optimize_rules = False
    train_pointset_acts_via_deltas = True
    ##### model parameters optimized from the MANO hand expanded set trajectory #####
    ckpt_fn = "ckpts/grab/102/dyn_mano_pointset_optimized_acts_optimized_ps_optimized_acts.pth"
    load_optimized_init_actions = "ckpts/grab/102/dyn_mano_pointset_optimized_acts_optimized_ps_optimized_acts.pth"
    load_optimized_init_transformations = "ckpts/grab/102/dyn_mano_shadow_arti_retar.pth"
    finger_cd_loss = 1.0
    optimize_pointset_motion_only = True
    ###### Retargeting Stage 1 ######



    ###### Retargeting Stage 2 ######
    load_optimized_init_transformations = "ckpts/grab/102/dyn_mano_shadow_arti_retar_optimized_arti.pth"
    ###### Retargeting Stage 2 ######




    use_opt_rigid_translations=True

    train_def = True
    optimizable_rigid_translations=True
    
    nerf {
        D = 8,
        d_in = 4,
        d_in_view = 3,
        W = 256,
        multires = 10,
        multires_view = 4,
        output_ch = 4,
        skips=[4],
        use_viewdirs=True
    }

    sdf_network {
        d_out = 257,
        d_in = 3,
        d_hidden = 256,
        n_layers = 8,
        skip_in = [4],
        multires = 6,
        bias = 0.5,
        scale = 1.0,
        geometric_init = True,
        weight_norm = True,
    } 

    variance_network {
        init_val = 0.3
    }

    rendering_network {
        d_feature = 256,
        mode = idr,
        d_in = 9,
        d_out = 3,
        d_hidden = 256,
        n_layers = 4,
        weight_norm = True,
        multires_view = 4,
        squeeze_out = True,
    }

    neus_renderer {
        n_samples = 64,
        n_importance = 64,
        n_outside = 0,
        up_sample_steps = 4 ,
        perturb = 1.0,
    }

    bending_network {
        multires = 6,
        bending_latent_size = 32,
        d_in = 3,
        rigidity_hidden_dimensions = 64,
        rigidity_network_depth = 5,
        use_rigidity_network = False,
        bending_n_timesteps = 10,
    }
}