package second.protos

Get desktop application:
View/edit binary Protocol Buffers messages

Configuration message for the AdamOptimizer See: https://www.tensorflow.org/api_docs/python/tf/train/AdamOptimizer

Used in: Optimizer

• optional LearningRate learning_rate = 1

• float weight_decay = 2

• bool amsgrad = 3

Used in: ClassSetting

• repeated float sizes = 1

• repeated float anchor_ranges = 2

• repeated float rotations = 3

• repeated float custom_values = 4

Used in: ClassSetting

• repeated float sizes = 1

• repeated float strides = 2

• repeated float offsets = 3

• repeated float rotations = 4

• repeated float custom_values = 5

Used in: BoxCoder

• bool linear_dim = 1

• bool encode_angle_vector = 2

• float z_fixed = 3

• float h_fixed = 4

Classification loss using a sigmoid function over the class prediction with the highest prediction score.

Used in: ClassificationLoss

• float alpha = 1

  Interpolation weight between 0 and 1.

• bool hard_bootstrap = 2

  Whether hard boot strapping should be used or not. If true, will only use one class favored by model. Othewise, will use all predicted class probabilities.

• bool anchorwise_output = 3

  DEPRECATED, do not use. Output loss per anchor.

Configuration proto for the box coder to be used in the object detection pipeline. See core/box_coder.py for details.

Used in: VoxelNet

• oneof box_coder

  • GroundBox3dCoder ground_box3d_coder = 1

  • BevBoxCoder bev_box_coder = 2

Used in: TargetAssigner

• oneof anchor_generator

  • AnchorGeneratorStride anchor_generator_stride = 1

  • AnchorGeneratorRange anchor_generator_range = 2

  • NoAnchor no_anchor = 3

• optional RegionSimilarityCalculator region_similarity_calculator = 4

• bool use_multi_class_nms = 5

• bool use_rotate_nms = 6

• int32 nms_pre_max_size = 7

• int32 nms_post_max_size = 8

• float nms_score_threshold = 9

• float nms_iou_threshold = 10

• float matched_threshold = 11

• float unmatched_threshold = 12

• string class_name = 13

• repeated int64 feature_map_size = 14

  3D zyx (DHW) size

Configuration for class prediction loss function.

Used in: Loss

• oneof classification_loss

  • WeightedSigmoidClassificationLoss weighted_sigmoid = 1

  • WeightedSoftmaxClassificationLoss weighted_softmax = 2

  • BootstrappedSigmoidClassificationLoss bootstrapped_sigmoid = 3

  • SigmoidFocalClassificationLoss weighted_sigmoid_focal = 4

  • SoftmaxFocalClassificationLoss weighted_softmax_focal = 5

  • GHMClassificationLoss weighted_ghm = 6

Used in: DatabasePreprocessingStep

• repeated int32 removed_difficulties = 1

Used in: DatabasePreprocessingStep

• map<string, uint32> min_num_point_pairs = 1

Used in: InputReader.Preprocess, Sampler

• oneof database_preprocessing_step

  • DBFilterByDifficulty filter_by_difficulty = 1

  • DBFilterByMinNumPointInGroundTruth filter_by_min_num_points = 2

Used in: TrainEvalPipelineConfig

• oneof model

  • VoxelNet second = 1

Configuration for intersection-over-union (IOU) similarity calculator.

Used in: RegionSimilarityCalculator

• float distance_norm = 1

• bool with_rotation = 2

• float rotation_alpha = 3

Used in: LearningRate

• float initial_learning_rate = 1

• float decay_length = 2

  must in range (0, 1)

• float decay_factor = 3

• bool staircase = 4

Used in: ClassificationLoss

• bool anchorwise_output = 1

• int32 bins = 2

• float momentum = 3

Used in: BoxCoder

• bool linear_dim = 1

• bool encode_angle_vector = 2

Used in: Sampler

• map<string, uint32> name_to_max_num = 1

Configuation for hard example miner.

Used in: Loss

• int32 num_hard_examples = 1

  Maximum number of hard examples to be selected per image (prior to enforcing max negative to positive ratio constraint). If set to 0, all examples obtained after NMS are considered.

• float iou_threshold = 2

  Minimum intersection over union for an example to be discarded during NMS.

• HardExampleMiner.LossType loss_type = 3

• int32 max_negatives_per_positive = 4

  Maximum number of negatives to retain for each positive anchor. If num_negatives_per_positive is 0 no prespecified negative:positive ratio is enforced.

• int32 min_negatives_per_image = 5

  Minimum number of negative anchors to sample for a given image. Setting this to a positive number samples negatives in an image without any positive anchors and thus not bias the model towards having at least one detection per image.

Whether to use classification losses ('cls', default), localization losses ('loc') or both losses ('both'). In the case of 'both', cls_loss_weight and loc_loss_weight are used to compute weighted sum of the two losses.

Used in: HardExampleMiner

• BOTH = 0

• CLASSIFICATION = 1

• LOCALIZATION = 2

Used in: TrainEvalPipelineConfig

• uint32 batch_size = 1

• optional InputReader.Dataset dataset = 2

• optional InputReader.Preprocess preprocess = 3

• uint32 max_num_epochs = 4

  deprecated

• uint32 prefetch_size = 5

  deprecated

Used in: InputReader

• string kitti_info_path = 1

• string kitti_root_path = 2

• string dataset_class_name = 3

  support KittiDataset and NuScenesDataset

Used in: InputReader

• bool shuffle_points = 1

• uint32 max_number_of_voxels = 2

• repeated float groundtruth_localization_noise_std = 3

• repeated float groundtruth_rotation_uniform_noise = 4

• repeated float global_rotation_uniform_noise = 5

• repeated float global_scaling_uniform_noise = 6

• repeated float global_translate_noise_std = 7

• bool remove_unknown_examples = 8

• uint32 num_workers = 9

• float anchor_area_threshold = 10

• bool remove_points_after_sample = 11

• float groundtruth_points_drop_percentage = 12

• uint32 groundtruth_drop_max_keep_points = 13

• bool remove_environment = 14

• repeated float global_random_rotation_range_per_object = 15

• repeated DatabasePreprocessingStep database_prep_steps = 16

• optional Sampler database_sampler = 17

• bool use_group_id = 18

  this will enable group sample and noise

• int64 min_num_of_points_in_gt = 19

  gt boxes contains less than this will be ignored.

• bool random_flip_x = 20

• bool random_flip_y = 21

• float sample_importance = 22

Used in: AdamOptimizer, MomentumOptimizer, RMSPropOptimizer

• oneof learning_rate

  • MultiPhase multi_phase = 1

  • OneCycle one_cycle = 2

  • ExponentialDecay exponential_decay = 3

  • ManualStepping manual_stepping = 4

Used in: MultiPhase

• float start = 1

• string lambda_func = 2

• string momentum_lambda_func = 3

Configuration for bounding box localization loss function.

Used in: Loss

• oneof localization_loss

  • WeightedL2LocalizationLoss weighted_l2 = 1

  • WeightedSmoothL1LocalizationLoss weighted_smooth_l1 = 2

  • WeightedGHMLocalizationLoss weighted_ghm = 3

• bool encode_rad_error_by_sin = 4

Message for configuring the localization loss, classification loss and hard example miner used for training object detection models. See core/losses.py for details

Used in: VoxelNet

• optional LocalizationLoss localization_loss = 1

  Localization loss to use.

• optional ClassificationLoss classification_loss = 2

  Classification loss to use.

• optional HardExampleMiner hard_example_miner = 3

  If not left to default, applies hard example mining.

• float classification_weight = 4

  Classification loss weight.

• float localization_weight = 5

  Localization loss weight.

Used in: LearningRate

• repeated float boundaries = 1

  must in range (0, 1)

• repeated float rates = 2

Configuration message for the MomentumOptimizer See: https://www.tensorflow.org/api_docs/python/tf/train/MomentumOptimizer

Used in: Optimizer

• optional LearningRate learning_rate = 1

• float momentum_optimizer_value = 2

• float weight_decay = 3

Used in: LearningRate

• repeated LearningRatePhase phases = 1

Configuration for intersection-over-union (IOU) similarity calculator.

Used in: RegionSimilarityCalculator

(message has no fields)

Used in: ClassSetting

(message has no fields)

Used in: LearningRate

• float lr_max = 1

• repeated float moms = 2

• float div_factor = 3

• float pct_start = 4

Top level optimizer message.

Used in: TrainConfig

• oneof optimizer

  • RMSPropOptimizer rms_prop_optimizer = 1

  • MomentumOptimizer momentum_optimizer = 2

  • AdamOptimizer adam_optimizer = 3

• bool use_moving_average = 4

• float moving_average_decay = 5

• bool fixed_weight_decay = 6

  i.e. AdamW

Configuration message for the RMSPropOptimizer See: https://www.tensorflow.org/api_docs/python/tf/train/RMSPropOptimizer

Used in: Optimizer

• optional LearningRate learning_rate = 1

• float momentum_optimizer_value = 2

• float decay = 3

• float epsilon = 4

• float weight_decay = 5

Configuration proto for region similarity calculators. See core/region_similarity_calculator.py for details.

Used in: ClassSetting

• oneof region_similarity

  • RotateIouSimilarity rotate_iou_similarity = 1

  • NearestIouSimilarity nearest_iou_similarity = 2

  • DistanceSimilarity distance_similarity = 3

Configuration for intersection-over-union (IOU) similarity calculator.

Used in: RegionSimilarityCalculator

(message has no fields)

Used in: InputReader.Preprocess

• string database_info_path = 1

• repeated Group sample_groups = 2

• repeated DatabasePreprocessingStep database_prep_steps = 3

• repeated float global_random_rotation_range_per_object = 4

• float rate = 5

Sigmoid Focal cross entropy loss as described in https://arxiv.org/abs/1708.02002

Used in: ClassificationLoss

• bool anchorwise_output = 1

  DEPRECATED, do not use.

• float gamma = 2

  modulating factor for the loss.

• float alpha = 3

  alpha weighting factor for the loss.

Sigmoid Focal cross entropy loss as described in https://arxiv.org/abs/1708.02002

Used in: ClassificationLoss

• bool anchorwise_output = 1

  DEPRECATED, do not use.

• float gamma = 2

  modulating factor for the loss.

• float alpha = 3

  alpha weighting factor for the loss.

Used in: VoxelNet

• repeated ClassSetting class_settings = 1

• float sample_positive_fraction = 2

• uint32 sample_size = 3

• bool assign_per_class = 4

• repeated int64 nms_pre_max_sizes = 5

  this will override setting in ClassSettings if provide.

• repeated int64 nms_post_max_sizes = 6

  this will override setting in ClassSettings if provide.

• repeated int64 nms_score_thresholds = 7

  this will override setting in ClassSettings if provide.

• repeated int64 nms_iou_thresholds = 8

  this will override setting in ClassSettings if provide.

Used in: TrainEvalPipelineConfig

• optional Optimizer optimizer = 1

• uint32 steps = 2

• uint32 steps_per_eval = 3

• uint32 save_checkpoints_secs = 4

• uint32 save_summary_steps = 5

• bool enable_mixed_precision = 6

• float loss_scale_factor = 7

• bool clear_metrics_every_epoch = 8

Convenience message for configuring a training and eval pipeline. Allows all of the pipeline parameters to be configured from one file.

• optional DetectionModel model = 1

• optional InputReader train_input_reader = 2

• optional TrainConfig train_config = 3

• optional InputReader eval_input_reader = 4

Used in: VoxelNet

• repeated float voxel_size = 1

• repeated float point_cloud_range = 2

• uint32 max_number_of_points_per_voxel = 3

• bool full_empty_part_with_mean = 4

• bool block_filtering = 5

• int64 block_factor = 6

• int64 block_size = 7

• float height_threshold = 8

Used in: DetectionModel

• string network_class_name = 1

• optional VoxelGenerator voxel_generator = 2

• optional VoxelNet.VoxelFeatureExtractor voxel_feature_extractor = 3

• optional VoxelNet.MiddleFeatureExtractor middle_feature_extractor = 4

• optional VoxelNet.RPN rpn = 5

• optional VoxelNet.RPN vpn = 25

• uint32 num_point_features = 6

• bool use_sigmoid_score = 7

• optional Loss loss = 8

• bool encode_rad_error_by_sin = 9

• bool encode_background_as_zeros = 10

• bool use_direction_classifier = 11

• float direction_loss_weight = 12

• float pos_class_weight = 13

• float neg_class_weight = 14

• VoxelNet.LossNormType loss_norm_type = 15

• optional BoxCoder box_coder = 16

• optional TargetAssigner target_assigner = 17

• repeated float post_center_limit_range = 18

• float direction_offset = 19

• float sin_error_factor = 20

• bool nms_class_agnostic = 21

• int64 num_direction_bins = 22

• float direction_limit_offset = 23

• bool lidar_input = 24

  deprecated in future

Used in: VoxelNet

• NormByNumExamples = 0

• NormByNumPositives = 1

• NormByNumPosNeg = 2

• DontNorm = 3

Used in: VoxelNet

• string module_class_name = 1

• repeated int32 num_filters_down1 = 2

• repeated int32 num_filters_down2 = 3

• int32 num_input_features = 4

• int32 downsample_factor = 5

Used in: VoxelNet

• string module_class_name = 1

• repeated int32 layer_nums = 2

• repeated int32 layer_strides = 3

• repeated int32 num_filters = 4

• repeated double upsample_strides = 5

• repeated int32 num_upsample_filters = 6

• bool use_groupnorm = 7

• int32 num_groups = 8

• int32 num_input_features = 9

Used in: VoxelNet

• string module_class_name = 1

• repeated int32 num_filters = 2

• bool with_distance = 3

• int32 num_input_features = 4

Used in: LocalizationLoss

• bool anchorwise_output = 1

  DEPRECATED, do not use. Output loss per anchor.

• float mu = 2

• int32 bins = 3

• float momentum = 4

• repeated float code_weight = 5

L2 location loss: 0.5 * ||weight * (a - b)|| ^ 2

Used in: LocalizationLoss

• bool anchorwise_output = 1

  DEPRECATED, do not use. Output loss per anchor.

• repeated float code_weight = 2

Classification loss using a sigmoid function over class predictions.

Used in: ClassificationLoss

• bool anchorwise_output = 1

  DEPRECATED, do not use. Output loss per anchor.

SmoothL1 (Huber) location loss: .5 * x ^ 2 if |x| < 1 else |x| - .5

Used in: LocalizationLoss

• bool anchorwise_output = 1

  DEPRECATED, do not use. Output loss per anchor.

• float sigma = 2

• repeated float code_weight = 3

Classification loss using a softmax function over class predictions.

Used in: ClassificationLoss

• bool anchorwise_output = 1

  DEPRECATED, do not use. Output loss per anchor.

• float logit_scale = 2

  Scale logit (input) value before calculating softmax classification loss. Typically used for softmax distillation.