package avod.protos

Get desktop application:
View/edit binary Protocol Buffers messages

Used in: LayersConfig

• oneof fc_layers

  • BasicFcLayers basic_fc_layers = 1

  • FusionFcLayers fusion_fc_layers = 2

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

Used in: Optimizer

• optional LearningRate learning_rate = 1

Used in: ModelConfig

• required int32 avod_proposal_roi_crop_size = 1

  AVOD Proposal ROI crop size

• required string avod_positive_selection = 3

  Positive selection, one of ['corr_cls', 'not_bkg']

• required int32 avod_nms_size = 4

  AVOD Non-max suppression boxes

• required float avod_nms_iou_thresh = 5

  AVOD NMS IoU threshold

• required string avod_box_representation = 6

  AVOD bounding box representation, one of ['box_3d', 'box_8c']

Used in: AVODLayersConfig

• required int32 num_layers = 1

• repeated int32 layer_sizes = 2

• required float l2_weight_decay = 3

  L2 weight decay

• required float keep_prob = 4

  Dropout keep probability

• required string fusion_method = 5

  Fusion method ('mean', 'concat')

Used in: KittiUtilsConfig

• oneof bev_maps_type

  • BevGenerator.Slices slices = 1

Used in: BevGenerator

• required float height_lo = 1

  Min and max height

• required float height_hi = 2

• required int32 num_slices = 3

  Number of slices to create

Configuration message for a constant learning rate.

Used in: LearningRate

• optional float learning_rate = 1

Message for configuring DetectionModel evaluator.

Used in: NetworkPipelineConfig

• optional uint32 eval_interval = 1

  Evaluation intervals during training

• optional string eval_mode = 2

  Evaluation mode, 'val' or 'test'

• repeated int32 ckpt_indices = 3

  Checkpoint indices to evaluate

• optional bool evaluate_repeatedly = 4

  Evaluate repeatedly while waiting for new checkpoints

• optional bool allow_gpu_mem_growth = 5

  GPU options

• optional float kitti_score_threshold = 6

  Kitti native evaluation

Configuration message for an exponentially decaying learning rate. See https://www.tensorflow.org/versions/master/api_docs/python/train/ \ decaying_the_learning_rate#exponential_decay

Used in: LearningRate

• optional float initial_learning_rate = 1

• optional uint32 decay_steps = 2

• optional float decay_factor = 3

• optional bool staircase = 4

Used in: LayersConfig

• oneof feature_extractor

  • VGGLayersConfig bev_vgg = 1

  • VGGLayersConfig img_vgg = 2

  • PyramidLayersConfig img_vgg_pyr = 3

  • PyramidLayersConfig bev_vgg_pyr = 4

Used in: AVODLayersConfig

• required int32 num_layers = 1

• repeated int32 layer_sizes = 2

• required float l2_weight_decay = 3

  L2 weight decay

• required float keep_prob = 4

  Dropout keep probability

• required string fusion_method = 5

  Fusion method ('mean', 'concat')

• required string fusion_type = 6

  Fusion type (early, late, deep)

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

Used in: Optimizer

• optional LearningRate learning_rate = 1

Used in: ModelConfig

• optional int32 bev_dims_h = 1

  Bev dimensions

• optional int32 bev_dims_w = 2

• optional int32 bev_depth = 3

• optional int32 img_dims_h = 4

  Image dimensions

• optional int32 img_dims_w = 5

• optional int32 img_depth = 6

Used in: NetworkPipelineConfig

• optional string name = 1

  Unique name for dataset

• optional string dataset_dir = 2

  Top level directory of the dataset

• optional string data_split = 3

  Split for the data (e.g. 'train', 'val')

• optional string data_split_dir = 4

  Folder that holds the data for the chosen data split

• optional bool has_labels = 5

  Whether the samples have labels

• optional string cluster_split = 6

  The data split to be used for calculating clusters (e.g. val split should use the train split for clustering)

• repeated string classes = 7

  Classes to be classified (e.g. ['Car', 'Pedestrian', 'Cyclist']

• repeated int32 num_clusters = 8

  Number of clusters corresponding to each class (e.g. [2, 1, 2])

• required string bev_source = 9

  BEV source, e.g. 'lidar'

• repeated string aug_list = 10

  Augmentations (e.g. [''], ['flipping'], ['flipping', 'pca_jitter'])

• optional KittiUtilsConfig kitti_utils_config = 20

  KittiUtils configuration

Used in: KittiDatasetConfig

• repeated float area_extents = 1

  3D area extents [min_x, max_x, min_y, max_y, min_z, max_z]

• required float voxel_size = 2

  Voxel grid size (for 2D and 3D)

• repeated float anchor_strides = 3

  Anchor strides

• optional int32 density_threshold = 4

  Anchor filtering density threshold

• required BevGenerator bev_generator = 20

• required MiniBatchConfig mini_batch_config = 21

Message for configuring Model Layer params.

Used in: ModelConfig

• required FeatureExtractor bev_feature_extractor = 1

• required FeatureExtractor img_feature_extractor = 2

• required RPNLayersConfig rpn_config = 3

• required AVODLayersConfig avod_config = 4

Configuration message for optimizer learning rate.

Used in: AdamOptimizer, GradientDescentOptimizer, MomentumOptimizer, RMSPropOptimizer

• oneof learning_rate

  • ConstantLearningRate constant_learning_rate = 1

  • ExponentialDecayLearningRate exponential_decay_learning_rate = 2

  • ManualStepLearningRate manual_step_learning_rate = 3

Used in: ModelConfig

• required float reg_loss_weight = 1

  RPN/AVOD Regression loss weight

• required float ang_loss_weight = 2

  AVOD angle vector loss weight

• required float cls_loss_weight = 3

  RPN/AVOD Classification loss weight

Configuration message for a manually defined learning rate schedule.

Used in: LearningRate

• optional float initial_learning_rate = 1

• repeated ManualStepLearningRate.LearningRateSchedule schedule = 2

Used in: ManualStepLearningRate

• optional uint32 step = 1

• optional float learning_rate = 2

Used in: MiniBatchConfig

• required MiniBatchIouThresholds iou_2d_thresholds = 1

  AVOD positive/negative 2D iou ranges

• required int32 mini_batch_size = 2

  Number of anchors in an AVOD mini batch

Used in: KittiUtilsConfig

• required int32 density_threshold = 1

  Density threshold for removing empty anchors

• required MiniBatchRpnConfig rpn_config = 2

• required MiniBatchAvodConfig avod_config = 3

Used in: MiniBatchAvodConfig, MiniBatchRpnConfig

• required float neg_iou_lo = 3

  RPN negative/positive iou ranges

• required float neg_iou_hi = 4

• required float pos_iou_lo = 5

• required float pos_iou_hi = 6

Used in: MiniBatchConfig

• oneof iou_type

  • MiniBatchIouThresholds iou_2d_thresholds = 1

  • MiniBatchIouThresholds iou_3d_thresholds = 2

• required int32 mini_batch_size = 3

  Number of anchors in an RPN mini batch

Message for configuring the DetectionModel.

Used in: NetworkPipelineConfig

• optional string model_name = 1

  Model name used to run either RPN or AVOD

• optional string checkpoint_name = 2

  Checkpoint name

• optional PathsConfig paths_config = 3

• required InputConfig input_config = 4

• required RpnConfig rpn_config = 5

• required AvodConfig avod_config = 6

• required float label_smoothing_epsilon = 7

  Label smoothing epsilon

• required float expand_proposals_xz = 8

  Expand proposals lengths along x and z for larger context region (in m) (0.0 - 1.0 recommended)

• repeated float path_drop_probabilities = 9

  Global path drop (p_keep_img, p_keep_bev) To disable path drop, set both to 1.0

• required bool train_on_all_samples = 10

  To keep all the samples including the ones without anchor-info i.e. labels during training

• required bool eval_all_samples = 11

  To keep all the samples including the ones without anchor-info i.e. labels during validation

• required LayersConfig layers_config = 12

  Layer configurations

• required LossConfig loss_config = 13

  Loss configurations

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

Used in: Optimizer

• optional LearningRate learning_rate = 1

• optional float momentum_optimizer_value = 2

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

• optional ModelConfig model_config = 1

  Detection Model config

• optional TrainConfig train_config = 2

  Training config

• optional EvalConfig eval_config = 3

  Evaluation config

• optional KittiDatasetConfig dataset_config = 4

  KittiDataset configuration

Top level optimizer message.

Used in: TrainConfig

• oneof optimizer

  • RMSPropOptimizer rms_prop_optimizer = 1

  • MomentumOptimizer momentum_optimizer = 2

  • AdamOptimizer adam_optimizer = 3

  • GradientDescentOptimizer gradient_descent = 6

• optional bool use_moving_average = 4

• optional float moving_average_decay = 5

Used in: ModelConfig

• optional string checkpoint_dir = 1

  Checkpoint dir

• optional string logdir = 2

  Log dir (no underscore to match tensorboard)

• optional string pred_dir = 3

  Directory to save predictions

Used in: FeatureExtractor

• repeated int32 vgg_conv1 = 1

  Conv layer 1 [repeat, num filter]

• repeated int32 vgg_conv2 = 2

  Conv layer 2 [repeat, num filter]

• repeated int32 vgg_conv3 = 3

  Conv layer 3 [repeat, num filter]

• repeated int32 vgg_conv4 = 4

  Conv layer 4 [repeat, num filter]

• optional float l2_weight_decay = 6

  L2 norm weight decay

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

Used in: Optimizer

• optional LearningRate learning_rate = 1

• optional float momentum_optimizer_value = 2

• optional float decay = 3

• optional float epsilon = 4

Used in: LayersConfig

• required int32 cls_fc6 = 1

  Anchor predictor layer configs classification fc layer size

• required int32 cls_fc7 = 2

• required int32 reg_fc6 = 3

  Regression fc layer size

• required int32 reg_fc7 = 4

• required float l2_weight_decay = 6

  L2 weight decay

• optional float keep_prob = 5

  Dropout probability - the probabilit that a neuron's output is kept during dropout

Used in: ModelConfig

• required int32 rpn_proposal_roi_crop_size = 1

  RPN proposal ROI crop size

• required string rpn_fusion_method = 2

  RPN proposal ROI fusion method, one of ['mean', 'concat']

• required int32 rpn_train_nms_size = 3

  RPN Non-max suppression boxes during training

• required int32 rpn_test_nms_size = 4

  RPN Non-max suppression boxes during testing

• required float rpn_nms_iou_thresh = 5

  RPN NMS IoU threshold

Message for configuring DetectionModel training jobs (train.py).

Used in: NetworkPipelineConfig

• optional uint32 batch_size = 1

  Input queue batch size.

• required uint32 max_iterations = 2

  Max training iteration

• optional Optimizer optimizer = 3

  Optimizer used to train the DetectionModel.

• optional uint32 checkpoint_interval = 4

  Checkpoint options

• optional uint32 max_checkpoints_to_keep = 5

• optional bool overwrite_checkpoints = 6

• required uint32 summary_interval = 7

  Summary options

• required bool summary_histograms = 8

• required bool summary_img_images = 9

• required bool summary_bev_images = 10

• optional bool allow_gpu_mem_growth = 11

  GPU options

Used in: FeatureExtractor

• repeated int32 vgg_conv1 = 1

  Conv layer 1 [repeat, num filter]

• repeated int32 vgg_conv2 = 2

  Conv layer 2 [repeat, num filter]

• repeated int32 vgg_conv3 = 3

  Conv layer 3 [repeat, num filter]

• repeated int32 vgg_conv4 = 4

  Conv layer 4 [repeat, num filter]

• required int32 upsampling_multiplier = 5

  Upsampling multiplier

• optional float l2_weight_decay = 6

  L2 norm weight decay