Default package

Get desktop application:
View/edit binary Protocol Buffers messages

Used in: Optimizer

• optional float initial_accumulator_value = 1

Used in: Optimizer

• optional float beta1 = 1

• optional float beta2 = 2

• optional float epsilon = 3

Configuration proto for batch norm to apply after convolution op. See https://www.tensorflow.org/api_docs/python/tf/contrib/layers/batch_norm

Used in: Hyperparams

• optional float decay = 1

• optional bool center = 2

• optional bool scale = 3

• optional float epsilon = 4

• optional bool train = 5

  Whether to train the batch norm variables. If this is set to false during training, the current value of the batch_norm variables are used for forward pass but they are never updated.

Used in: Cap2SGPreprocess

• BIAS_MODE_ZERO = 1

• BIAS_MODE_TRADITION = 2

• BIAS_MODE_TRAIN_FROM_EMBEDDING = 3

• optional Cap2SGPreprocess preprocess_options = 1

  Preprocess options.

• optional Cap2SGLinguistic linguistic_options = 33

• optional Cap2SGGrounding grounding_options = 2

• optional Cap2SGDetection detection_options = 3

• optional Cap2SGRelation relation_options = 4

• optional Cap2SGCommonSense common_sense_options = 5

• optional bool parse_attribute_in_training = 6

• optional bool parse_attribute_in_evaluation = 7

• optional float weight_decay = 8

• optional float dropout_keep_prob = 9

• optional uint32 hidden_units = 12

• optional Transformer transformer_config = 13

  If true, use self-attention.

• optional bool common_sense_refine = 14

Used in: Cap2SG

• optional float loss_weight = 1

  Relation loss weight.

• optional float triple_loss_weight = 7

• optional uint32 beam_size = 6

• optional uint32 relation_max_total_size = 2

• optional float rnn_output_keep_prob = 3

• optional float rnn_input_keep_prob = 4

• optional float rnn_state_keep_prob = 5

• optional bool update_gradients = 8

• optional uint32 rnn_hidden_units = 11

• optional RelationScoringMethod relation_scoring_method = 12

• optional float ranking_score_weight = 13

Used in: Cap2SG

• optional float loss_weight = 1

  Detection loss weight.

• optional PostProcess post_process = 2

  Post processing config.

• optional bool override_grounding_result = 3

  If true, override the grounding output. Required for iterative refinement.

• optional DetectionFeatureType detection_feature_type = 4

  Detection feature type.

• optional uint32 num_iterations = 5

  Number of refining iterations.

• optional float grounding_iou_threshold = 6

• optional bool used_detection_head_as_rnn_vis_feature = 13

• optional bool predict_attributes = 14

Used in: Cap2SG

• optional float loss_weight = 1

  Grounding loss weight.

• optional float midn_attention_dropout_keep_prob = 8

• optional bool two_branches = 9

Used in: Cap2SG

• optional float weight_decay = 41

  Configuation of the graph network.

• optional GraphNetwork graph_network = 42

Used in: Cap2SG

• optional string vocabulary_file = 1

  Vocabulary file.

• optional string closed_vocabulary_file = 2

  Closed vocabulary file.

• optional uint32 minimum_frequency = 3

  Minimum frequency to incorporate the word in the vocabulary.

• optional uint32 embedding_dims = 4

  Dimensions of the word embedding vectors.

• optional bool embedding_trainable = 5

  If true, train the word embedding vectors.

• optional string glove_vocabulary_file = 6

  GloVe word tokens.

• optional string glove_embedding_file = 7

  GloVe word embeddings for initializing the word embedding vectors.

• optional BiasMode bias_mode = 12

  Mode for computing the bias.

Used in: Cap2SG

• optional float loss_weight = 1

  Relation loss weight.

• optional uint32 relation_max_total_size = 2

• optional uint32 relation_max_size_per_class = 3

• optional float relation_threshold = 4

• optional float grounding_iou_threshold = 6

Used in: Reader

• repeated string input_pattern = 1

  Pattern of the input files.

• optional int32 shuffle_buffer_size = 2

  Shuffle buffer size.

• optional int32 interleave_cycle_length = 3

  Interleave cycle length.

• optional int32 num_parallel_calls = 5

  Number of parallel calls.

• optional int32 batch_size = 6

  Batch size.

• optional int32 prefetch_buffer_size = 7

  Prefetch buffer size.

• optional int32 feature_dimensions = 8

  Dimensions of the pre-extracted FRCNN features.

• optional int32 max_n_proposal = 9

  Maximum number of proposals.

• optional int32 graph_index = 10

  For DEBUG purpose, use a fixed graph instead of sample one.

• optional uint32 num_hidden_layers = 1

  Number of hidden layers.

• optional uint32 num_attention_heads = 2

  Number of attention heads.

• optional uint32 intermediate_size = 3

  Size of the intermediate layer.

Used in: Cap2SGDetection

• DET_FEATURE_DEFAULT = 1

• DET_FEATURE_CNN = 2

• DET_FEATURE_ADD_ATTRIBUTE = 3

Used in: WSSGGenLing, WSSceneGraph, WSSceneGraphCaptionGNet, WSSceneGraphGNet

• ATTENTION = 1

• CLASSIFICATION = 2

• ATTENTION_x_CLASSIFICATION = 3

Used in: Pipeline

• optional uint32 steps = 1

  Positive number of steps for which to evaluate model. if `None`, evaluate util `input_fn` raises an end-of-input exception.

• optional uint32 start_delay_secs = 2

  Start evaluating after waiting for this many seconds.

• optional uint32 throttle_secs = 3

  Do not re-evaluate unless the last evaluation was started at least this many seconds ago.

Used in: LearningRateSchedule

• optional float initial_learning_rate = 1

• optional int32 decay_steps = 2

• optional float decay_rate = 3

• optional bool staircase = 4

FastRCNN configs.

• optional FasterRcnnFeatureExtractor feature_extractor = 1

  Feature extractor config.

• optional bool inplace_batchnorm_update = 2

  Whether to update batch_norm inplace during training. This is required for batch norm to work correctly on TPUs. When this is false, user must add a control dependency on tf.GraphKeys.UPDATE_OPS for train/loss op in order to update the batch norm moving average parameters.

• optional int32 initial_crop_size = 3

  Output size (width and height are set to be the same) of the initial bilinear interpolation based cropping during ROI pooling.

• optional int32 maxpool_kernel_size = 4

  Kernel size of the max pool op on the cropped feature map during ROI pooling.

• optional int32 maxpool_stride = 5

  Stride of the max pool op on the cropped feature map during ROI pooling.

• optional float dropout_keep_prob = 6

  Keep probability of the dropout layer.

• optional bool dropout_on_feature_map = 7

  If true, add a dropout layer after extracting feature map.

• optional string checkpoint_path = 8

  Path to the pre-trained checkpoint.

• optional bool from_classification_checkpoint = 9

  If true, initialize the FastRCNN model from a classification checkpoint.

• optional float weight_decay = 10

  Weight decay of the FastRCNN feature extractor.

Used in: FastRCNN

• optional string type = 1

  Type of Faster R-CNN model (e.g., 'faster_rcnn_resnet101'; See builders/model_builder.py for expected types).

• optional int32 first_stage_features_stride = 2

  Output stride of extracted RPN feature map.

• optional bool batch_norm_trainable = 3

  Whether to update batch norm parameters during training or not. When training with a relative large batch size (e.g. 8), it could be desirable to enable batch norm update.

Used in: Initializer

(message has no fields)

Used in: Initializer

(message has no fields)

Used in: Cap2SGLinguistic, WSSGGenLing, WSSceneGraphCaptionGNet, WSSceneGraphGNet

• oneof graph_network_oneof

  • NoGraph no_graph = 1

  • SelfAttention self_attention = 2

  • MessagePassing message_passing = 4

Configuration proto for the convolution op hyperparameters. See tensorflow/models/research/object_detection/protos/hyperparams.proto for details.

Used in: WSSGGenLing, WSSceneGraph, WSSceneGraphCaptionGNet, WSSceneGraphGNet

• optional Hyperparams.Op op = 1

• optional Regularizer regularizer = 2

  Regularizer for the weights of the convolution op.

• optional Initializer initializer = 3

  Initializer for the weights of the convolution op.

• optional Hyperparams.Activation activation = 4

• optional BatchNorm batch_norm = 5

  BatchNorm hyperparameters. If this parameter is NOT set then BatchNorm is not applied!

• optional bool regularize_depthwise = 6

  Whether depthwise convolutions should be regularized. If this parameter is NOT set then the conv hyperparams will default to the parent scope.

Type of activation to apply after convolution.

Used in: Hyperparams

• NONE = 0

  Use None (no activation)

• RELU = 1

  Use tf.nn.relu

• RELU_6 = 2

  Use tf.nn.relu6

Operations affected by hyperparameters.

Used in: Hyperparams

• CONV = 1

  Convolution, Separable Convolution, Convolution transpose.

• FC = 2

  Fully connected

Proto with one-of field for initializers.

Used in: Hyperparams

• oneof initializer_oneof

  • TruncatedNormalInitializer truncated_normal_initializer = 1

  • VarianceScalingInitializer variance_scaling_initializer = 2

  • RandomNormalInitializer random_normal_initializer = 3

  • GlorotNormalInitializer glorot_normal_initializer = 4

  • GlorotUniformInitializer glorot_uniform_initializer = 5

Configuration proto for L1 Regularizer. See https://www.tensorflow.org/api_docs/python/tf/contrib/layers/l1_regularizer

Used in: Regularizer

• optional float weight = 1

Configuration proto for L2 Regularizer. See https://www.tensorflow.org/api_docs/python/tf/contrib/layers/l2_regularizer

Used in: Regularizer

• optional float weight = 1

Used in: TrainConfig

• oneof learning_rate_schedule

  • PiecewiseConstantDecay piecewise_constant_decay = 1

  • ExponentialDecay exponential_decay = 2

  • PolynomialDecay polynomial_decay = 3

• LOSS_TYPE_SOFTMAX_CROSSENTROPY = 1

• LOSS_TYPE_SIGMOID_CROSSENTROPY = 2

• LOSS_TYPE_FOCAL_SIGMOID_CROSSENTROPY = 3

Used in: GraphNetwork

• optional uint32 n_layer = 1

  Number of stacked layers.

• optional uint32 n_head = 2

  Number of attention heads.

• optional uint32 key_dims = 3

  Dimensions of the queries and keys.

• optional bool add_bi_directional_edges = 6

  If true, add bi-directional edges object-predicate-subject.

• optional bool add_self_loop_edges = 7

  If true, add self-loop.

• optional bool use_reverse_edges = 10

• optional uint32 hidden_size = 8

• optional float attn_scale = 9

• optional float attn_dropout_keep_prob = 11

• optional float rnn_dropout_keep_prob = 12

Used in: Pipeline

(message has no fields)

A graph that directly returns the node/edge embeddings.

Used in: GraphNetwork

(message has no fields)

Used in: TrainConfig

• oneof optimizer

  • AdagradOptimizer adagrad = 1

  • RMSPropOptimizer rmsprop = 2

  • AdamOptimizer adam = 3

Used in: LearningRateSchedule

• repeated int32 boundaries = 1

• repeated float values = 2

• optional Reader train_reader = 1

  Train reader.

• optional Reader eval_reader = 2

  Eval reader.

• optional Reader test_reader = 3

  Test reader.

• optional Model model = 4

  Model config.

• optional TrainConfig train_config = 5

  Train config.

• optional EvalConfig eval_config = 6

  Eval config

• optional float seed = 7

  Random seed.

Used in: LearningRateSchedule

• optional float initial_learning_rate = 1

• optional int32 decay_steps = 2

• optional float end_learning_rate = 3

• optional float power = 4

• optional bool cycle = 5

Used in: Cap2SGDetection, WSSGGenLing, WSSceneGraph, WSSceneGraphCaptionGNet, WSSceneGraphGNet

• optional float score_thresh = 1

  Score threshhold for NMS.

• optional float iou_thresh = 2

  IoU threshold to check the overlap.

• optional int32 max_size_per_class = 3

  Maximum detections per class.

• optional int32 max_total_size = 4

  Maximum total detections.

Used in: Optimizer

• optional float decay = 1

• optional float momentum = 2

• optional float epsilon = 3

• optional bool centered = 4

Configuration proto for random normal initializer. See https://www.tensorflow.org/api_docs/python/tf/random_normal_initializer

Used in: Initializer

• optional float mean = 1

• optional float stddev = 2

Used in: Pipeline

• oneof reader_oneof

  • CaptionGraphReader caption_graph_reader = 5

Proto with one-of field for regularizers.

Used in: Hyperparams

• oneof regularizer_oneof

  • L1Regularizer l1_regularizer = 1

  • L2Regularizer l2_regularizer = 2

Used in: WSSceneGraphRnnRefine, WSSceneGraphRnnRefineV2

• ZEROS = 1

• POINTWISE_ADD = 2

• POINTWISE_MULT = 3

• SPATIAL = 4

• SPATIAL_POINTWISE_ADD = 5

Used in: Cap2SGCommonSense

• REL_LANGUAGE_SCORE_ONLY = 1

• REL_RANKING_SCORE_ONLY = 2

• REL_MIXED_SCORE = 3

A graph using self attention to update node embeddings.

Used in: GraphNetwork

• optional uint32 n_layer = 1

  Number of stacked layers.

• optional uint32 n_head = 2

  Number of attention heads.

• optional uint32 key_dims = 3

  Dimensions of the queries and keys.

• optional bool add_bi_directional_edges = 6

  If true, add bi-directional edges object-predicate-subject.

• optional bool add_self_loop_edges = 7

  If true, add self-loop.

Used in: Pipeline

• optional LearningRateSchedule learning_rate_schedule = 1

  Learning rate schedule.

• optional Optimizer optimizer = 2

  Optimizer to use.

• optional uint32 max_steps = 3

  Positive number of total steps for which to train model.

• optional int32 log_step_count_steps = 4

  The frequency, in number of global steps, that the global step/sec and the loss will be logged during training.

• optional uint32 save_summary_steps = 5

  Save summaries every this many steps.

• optional uint32 save_checkpoints_steps = 6

  Save checkpoints every this many steps.

• optional uint32 keep_checkpoint_max = 7

  The maximum number of recent checkpoint files to keep.

• optional float max_gradient_norm = 8

  Max gradient norm to be clipped.

Used in: Cap2SG

• optional uint32 num_hidden_layers = 2

  Nubmer of hidden layers.

• optional uint32 num_attention_heads = 3

  Number of attention heads.

• optional uint32 intermediate_size = 4

  Size of the intermediate layer.

• optional float attention_probs_dropout_prob = 5

  Attention dropout rate.

• optional float hidden_dropout_prob = 6

  Hidden dropout rate.

Configuration proto for truncated normal initializer. See https://www.tensorflow.org/api_docs/python/tf/truncated_normal_initializer

Used in: Initializer

• optional float mean = 1

• optional float stddev = 2

Configuration proto for variance scaling initializer. See https://www.tensorflow.org/api_docs/python/tf/contrib/layers/ variance_scaling_initializer

Used in: Initializer

• optional float factor = 1

• optional bool uniform = 2

• optional VarianceScalingInitializer.Mode mode = 3

Used in: VarianceScalingInitializer

• FAN_IN = 0

• FAN_OUT = 1

• FAN_AVG = 2

Used in: WSSceneGraphRnnRefine, WSSceneGraphRnnRefineV2

• optional string glove_vocab_file = 51

• optional string glove_embedding_file = 52

• optional string token_to_id_meta_file = 1

  Meta file containing the token to id mapping.

• optional string entity_emb_npy_file = 2

  Entity embedding file, data shape=[1 + #entities, dims].

• optional string predicate_emb_npy_file = 3

  Predicate embedding file, data shape=[1 + #predicates, dims].

• optional float dropout_keep_prob = 4

  Keep probability of dropout layers.

• optional float attn_dropout_keep_prob = 5

  Keep probability of attention dropout layers.

• optional float attn_scale = 10

  Scale factor of attention.

• optional Hyperparams fc_hyperparams = 6

  Hyperparameters for the fully-connected layers.

• optional uint32 entity_hidden_units = 7

  Entity: number of hidden units.

• optional uint32 relation_hidden_units = 8

  Relation: number of hidden units.

• optional float iou_threshold = 9

  IoU threshold to compute metrics.

• optional EdgeScoring edge_scoring = 15

  Edge scoring method.

• optional PostProcess post_process = 16

  Post processing.

• optional float iou_threshold_to_propogate = 17

  IoU threshold to propogate annotations.

• optional float iou_threshold_to_propogate_relation = 19

  IoU threshold to propogate annotations.

• optional uint32 n_refine_iteration = 18

  Refine iterations.

• optional bool refine_use_softmax = 24

  If true, the refine process uses sigmoid prediction.

• optional float multiple_entity_detection_loss_weight = 20

  Weight of the MED loss.

• optional float proposal_refine_loss_weight = 23

  Weight of the refine loss.

• optional float relation_refine_loss_weight = 28

  Weight of the refine loss.

• optional bool use_spatial_feature = 29

  If true, use spatial features.

• optional float joint_inferring_relation_weight = 32

  Weight of the proposal-proposal relation edge.

• optional float mps_relation_dropout_keep_prob = 33

  Dropout keep probability for the relation edges.

• optional bool use_log_prob = 34

  If true, use log probability.

• optional GraphNetwork text_graph_network = 42

  Configuation of the graph network.

• optional uint32 sage_steps = 43

  Sage steps, the (i+1) th refinement will starts `sage_steps` later than the (i) th refinement.

• optional bool train_graph_initial_embedding = 44

  If true, make the graph initial embedding trainable.

• optional float graph_initial_embedding_max_norm = 45

  If set, the max-norm of the graph initial embedding.

• optional string token_to_id_meta_file = 1

  Meta file containing the token to id mapping.

• optional string entity_emb_npy_file = 2

  Entity embedding file, data shape=[1 + #entities, dims].

• optional string predicate_emb_npy_file = 3

  Predicate embedding file, data shape=[1 + #predicates, dims].

• optional float dropout_keep_prob = 4

  Keep probability of dropout layers.

• optional float attn_dropout_keep_prob = 5

  Keep probability of attention dropout layers.

• optional float attn_scale = 10

  Scale factor of attention.

• optional Hyperparams fc_hyperparams = 6

  Hyperparameters for the fully-connected layers.

• optional uint32 entity_hidden_units = 7

  Entity: number of hidden units.

• optional uint32 relation_hidden_units = 8

  Relation: number of hidden units.

• optional float iou_threshold = 9

  IoU threshold to compute metrics.

• optional EdgeScoring edge_scoring = 15

  Edge scoring method.

• optional PostProcess post_process = 16

  Post processing.

• optional float iou_threshold_to_propogate = 17

  IoU threshold to propogate annotations.

• optional float iou_threshold_to_propogate_relation = 19

  IoU threshold to propogate annotations.

• optional uint32 n_refine_iteration = 18

  Refine iterations.

• optional bool refine_use_softmax = 24

  If true, the refine process uses sigmoid prediction.

• optional float multiple_entity_detection_loss_weight = 20

  Weight of the MED loss.

• optional float proposal_refine_loss_weight = 23

  Weight of the refine loss.

• optional float relation_refine_loss_weight = 28

  Weight of the refine loss.

• optional bool use_spatial_feature = 29

  If true, use spatial features.

• optional float joint_inferring_relation_weight = 32

  Weight of the proposal-proposal relation edge.

• optional float mps_relation_dropout_keep_prob = 33

  Dropout keep probability for the relation edges.

• optional bool use_log_prob = 34

  If true, use log probability.

Used in: WSSceneGraphRnnRefine, WSSceneGraphRnnRefineV2

• optional string glove_vocab_file = 51

• optional string glove_embedding_file = 52

• optional string token_to_id_meta_file = 1

  Meta file containing the token to id mapping.

• optional string entity_emb_npy_file = 2

  Entity embedding file, data shape=[1 + #entities, dims].

• optional string predicate_emb_npy_file = 3

  Predicate embedding file, data shape=[1 + #predicates, dims].

• optional float dropout_keep_prob = 4

  Keep probability of dropout layers.

• optional float attn_dropout_keep_prob = 5

  Keep probability of attention dropout layers.

• optional float attn_scale = 10

  Scale factor of attention.

• optional Hyperparams fc_hyperparams = 6

  Hyperparameters for the fully-connected layers.

• optional uint32 entity_hidden_units = 7

  Entity: number of hidden units.

• optional uint32 relation_hidden_units = 8

  Relation: number of hidden units.

• optional float iou_threshold = 9

  IoU threshold to compute metrics.

• optional EdgeScoring edge_scoring = 15

  Edge scoring method.

• optional PostProcess post_process = 16

  Post processing.

• optional float iou_threshold_to_propogate = 17

  IoU threshold to propogate annotations.

• optional float iou_threshold_to_propogate_relation = 19

  IoU threshold to propogate annotations.

• optional uint32 n_refine_iteration = 18

  Refine iterations.

• optional bool refine_use_softmax = 24

  If true, the refine process uses sigmoid prediction.

• optional float multiple_entity_detection_loss_weight = 20

  Weight of the MED loss.

• optional float proposal_refine_loss_weight = 23

  Weight of the refine loss.

• optional float relation_refine_loss_weight = 28

  Weight of the refine loss.

• optional bool use_spatial_feature = 29

  If true, use spatial features.

• optional float joint_inferring_relation_weight = 32

  Weight of the proposal-proposal relation edge.

• optional float mps_relation_dropout_keep_prob = 33

  Dropout keep probability for the relation edges.

• optional bool use_log_prob = 34

  If true, use log probability.

• optional GraphNetwork text_graph_network = 42

  Configuation of the graph network.

• optional uint32 sage_steps = 43

  Sage steps, the (i+1) th refinement will starts `sage_steps` later than the (i) th refinement.

• optional bool train_graph_initial_embedding = 44

  If true, make the graph initial embedding trainable.

• optional float graph_initial_embedding_max_norm = 45

  If set, the max-norm of the graph initial embedding.

Used in: WSSceneGraphRnnRefine, WSSceneGraphRnnRefineV2

• optional string glove_vocab_file = 51

• optional string glove_embedding_file = 52

• optional string token_to_id_meta_file = 1

  Meta file containing the token to id mapping.

• optional string entity_emb_npy_file = 2

  Entity embedding file, data shape=[1 + #entities, dims].

• optional string predicate_emb_npy_file = 3

  Predicate embedding file, data shape=[1 + #predicates, dims].

• optional float dropout_keep_prob = 4

  Keep probability of dropout layers.

• optional float attn_dropout_keep_prob = 5

  Keep probability of attention dropout layers.

• optional float attn_scale = 10

  Scale factor of attention.

• optional Hyperparams fc_hyperparams = 6

  Hyperparameters for the fully-connected layers.

• optional uint32 entity_hidden_units = 7

  Entity: number of hidden units.

• optional uint32 relation_hidden_units = 8

  Relation: number of hidden units.

• optional float iou_threshold = 9

  IoU threshold to compute metrics.

• optional EdgeScoring edge_scoring = 15

  Edge scoring method.

• optional PostProcess post_process = 16

  Post processing.

• optional float iou_threshold_to_propogate = 17

  IoU threshold to propogate annotations.

• optional float iou_threshold_to_propogate_relation = 19

  IoU threshold to propogate annotations.

• optional uint32 n_refine_iteration = 18

  Refine iterations.

• optional bool refine_use_softmax = 24

  If true, the refine process uses sigmoid prediction.

• optional float multiple_entity_detection_loss_weight = 20

  Weight of the MED loss.

• optional float proposal_refine_loss_weight = 23

  Weight of the refine loss.

• optional float relation_refine_loss_weight = 28

  Weight of the refine loss.

• optional bool use_spatial_feature = 29

  If true, use spatial features.

• optional float joint_inferring_relation_weight = 32

  Weight of the proposal-proposal relation edge.

• optional float mps_relation_dropout_keep_prob = 33

  Dropout keep probability for the relation edges.

• optional bool use_log_prob = 34

  If true, use log probability.

• optional GraphNetwork text_graph_network = 42

  Configuation of the graph network.

• optional uint32 sage_steps = 43

  Sage steps, the (i+1) th refinement will starts `sage_steps` later than the (i) th refinement.

• optional bool train_graph_initial_embedding = 44

  If true, make the graph initial embedding trainable.

• optional float graph_initial_embedding_max_norm = 45

  If set, the max-norm of the graph initial embedding.

• oneof proposasl_network_oneof

  • WSSceneGraphGNet ws_scene_graph_gnet = 1

  • WSSceneGraphCaptionGNet ws_scene_graph_caption_gnet = 2

  • WSSGGenLing ws_sggen_ling = 3

• optional uint32 rnn_layers = 34

  Number of RNN layers to use.

• optional uint32 rnn_hidden_units = 35

  Number of RNN hidden units.

• optional float rnn_input_keep_prob = 36

  Keep probability of RNN inputs.

• optional float rnn_output_keep_prob = 37

  Keep probability of RNN outputs.

• optional float rnn_state_keep_prob = 38

  Keep probability of RNN states.

• optional uint32 beam_size = 39

  Beam size for searching the solution.

• optional uint32 max_total_size = 40

  Maximum triples to retain.

• optional bool use_transformer = 42

  If true, use transformer to contextualize triple features.

• optional uint32 transformer_layers = 43

  Number of Transformer layers.

• optional float transformer_attn_dropout_prob = 44

  Dropout probability of the Transformer attention layers.

• optional RelationFeatureType relation_feature_type = 45

  Relation feature type.

• oneof proposasl_network_oneof

  • WSSceneGraphGNet ws_scene_graph_gnet = 1

  • WSSceneGraphCaptionGNet ws_scene_graph_caption_gnet = 2

  • WSSGGenLing ws_sggen_ling = 3

• optional uint32 rnn_layers = 34

  Number of RNN layers to use.

• optional uint32 rnn_hidden_units = 35

  Number of RNN hidden units.

• optional float rnn_input_keep_prob = 36

  Keep probability of RNN inputs.

• optional float rnn_output_keep_prob = 37

  Keep probability of RNN outputs.

• optional float rnn_state_keep_prob = 38

  Keep probability of RNN states.

• optional uint32 beam_size = 39

  Beam size for searching the solution.

• optional uint32 max_total_size = 40

  Maximum triples to retain.

• optional bool use_transformer = 42

  If true, use transformer to contextualize triple features.

• optional uint32 transformer_layers = 43

  Number of Transformer layers.

• optional float transformer_attn_dropout_prob = 44

  Dropout probability of the Transformer attention layers.

• optional RelationFeatureType relation_feature_type = 45

  Relation feature type.