config.json

{
  "architectures": [
    "MotionPredictorForRectifiedFlow"
  ],
  "camera_motion_conditioning_drop_prob": 0.1,
  "consistency_delta_time": 0.001,
  "consistency_loss_weight": 1.0,
  "consistency_velocity_match_alpha": 1e-05,
  "continuous_predictor": true,
  "default_track_rate": 30,
  "denoising_predictor_config": {
    "attention_bias": true,
    "attention_dropout": 0.0,
    "head_dim": 64,
    "hidden_act": "silu",
    "hidden_size": 768,
    "initializer_range": 0.02,
    "intermediate_size": 2046,
    "mlp_bias": true,
    "model_type": "spatial_track",
    "num_attention_heads": 12,
    "num_hidden_layers": 4,
    "num_key_value_heads": 12,
    "rms_norm_eps": 1e-06,
    "track_dimensionality": 2
  },
  "depth_track_multiplier": 1.0,
  "encoder_config": null,
  "encoder_pooler_type": "none",
  "extra_features_size": 3,
  "feature_map_upsample_method": "nearest",
  "frame_sample_rate": null,
  "freeze_global_image_model": true,
  "freeze_image_model": true,
  "freeze_text_encoder": true,
  "global_image_model_name": "openai/clip-vit-base-patch32",
  "global_video_encoder_hidden_size": 768,
  "height": 256,
  "image_model_batch_size": null,
  "image_model_name": "facebook/sam2.1-hiera-tiny",
  "late_noise_conditioning": true,
  "mask_non_visible_tracks": false,
  "max_height_shift": null,
  "max_track_length": 50,
  "max_width_shift": null,
  "model_type": "motion_predictor",
  "movement_weighting_loss_temperature": null,
  "movement_weighting_temperature": 0.5,
  "noise_schedule_type": "cos",
  "num_global_image_tokens": 49,
  "position_encoding_norm_to_feature_norm_ratio": 1.0,
  "prepend_query_points": true,
  "rectified_flow_ema_beta": 0.9999,
  "rectified_flow_ema_update_every": 100,
  "rectified_flow_relative_prediction": true,
  "rectified_flow_use_consistency": true,
  "sam2_image_size": 512,
  "text_conditioning_drop_prob": 0.1,
  "text_encoder_max_seq_length": 32,
  "text_encoder_name": "openai/clip-vit-base-patch32",
  "timestep_exponential_decay_loss_factor": null,
  "torch_dtype": "float32",
  "track_dimensionality": 2,
  "track_predictor_config": {
    "attention_bias": true,
    "attention_dropout": 0.0,
    "head_dim": 64,
    "hidden_act": "silu",
    "hidden_size": 768,
    "initializer_range": 0.02,
    "intermediate_size": 2046,
    "max_position_embeddings": 1024,
    "mlp_bias": true,
    "model_type": "joint_track",
    "num_attention_heads": 12,
    "num_hidden_layers": 12,
    "num_key_value_heads": 12,
    "rms_norm_eps": 1e-06,
    "rope_scaling": null,
    "rope_theta": 10000.0,
    "use_cache": true
  },
  "track_predictor_has_spatial_condition": true,
  "track_predictor_head_kwargs": {
    "concatenate_original_position": true,
    "norm_feature": true,
    "predictor_type": "linear"
  },
  "track_predictor_head_type": "denoising_predictor",
  "track_rate_conditioning_drop_prob": 0.0,
  "track_rate_conditioning_max_track_rate": 60,
  "track_subsample_count": 100,
  "transformers_version": "4.50.3",
  "use_absolute_positional_embeddings": false,
  "use_camera_motion_conditioning": true,
  "use_previous_relative_shift": true,
  "use_previous_relative_shift_input": false,
  "use_track_rate_conditioning": false,
  "video_conditioning_drop_prob": 0.0,
  "video_encoder_hidden_size": 352,
  "visible_min_ratio": 0.5,
  "width": 256
}