package fewshot.configs
Get desktop application:
View/edit binary Protocol Buffers messages
message
c4_config.proto:7Used in:
optional int32 height = 1
optional int32 width = 2
optional int32 num_channels = 3
repeated int32 num_filters = 4
optional float weight_decay = 5
optional string data_format = 6
optional string normalization = 7
repeated int32 num_groups = 8
Group norm parameters.
repeated int32 pool = 9
optional bool add_last_relu = 10
optional string pool_padding = 11
optional float activation_scaling = 12
optional bool add_dropout = 13
Dropout features.
optional float dropout_rate = 14
Dropout rate.
repeated int32 num_fc_dim = 15
message
environment_config.proto:6optional string data_folder = 1
Path to the dataset.
optional string dataset = 2
Dataset name.
optional string results = 3
Save folder.
optional string checkpoint = 4
Checkpoint folder (if different from save folder).
optional string train_split = 100
Split name for train.
optional string val_split = 101
Split name for validation.
optional string test_split = 102
Split name for test.
optional string train_fs_split = 103
Split name for few-shot training.
optional string val_fs_split = 104
Split name for few-shot validationn.
optional string test_fs_split = 105
Split name for few-shot test.
optional int32 image_size = 200
Image resize size.
optional int32 crop_size = 201
Image crop size.
optional bool random_crop = 202
Whether perform random crop.
optional bool random_flip = 203
Whether perform random flip.
optional bool random_color = 204
Whether perform random color.
optional bool random_rotate = 205
Whether perform random rotate.
repeated float mean_pix = 300
Mean pixel values for each channel.
repeated float std_pix = 301
Std pixel values for each channel.
message
episode_config.proto:5optional string base_sampler = 1
Base data sampler. Choice of `fewshot`, `incremental`, `constant_prob`, `crp`.
optional int32 nway = 2
Number of classes
optional int32 nquery = 3
Number of query examples per class.
optional int32 nshot_max = 4
Maximum shots for incremental sampler.
optional int32 max_num_per_cls = 5
Maximum number of support images per class.
optional bool semisupervised = 6
Whether an episode is supervised or not.
optional float label_ratio = 7
Target label ratio for semisupervised episodes.
optional int32 maxlen = 8
Maximum episode length.
optional bool allow_repeat = 9
Whether to allow repeating images in an episode.
optional bool use_query = 10
Whether to enable query set in an episode.
optional bool fix_unknown = 11
Whether to fix the unknown class ID.
optional bool hierarchical = 100
Whether making it a hierarchical sampler.
optional bool use_class_hierarchy = 101
Whether to use dataset class hierarchy.
optional int32 nstage = 102
Number of sequential stages for hierarchical sampler.
optional string blender = 103
Blender for blending multiple episodes.
optional int32 blur_window_size = 104
Blur blender window size.
optional int32 blur_stride = 105
Blur blender stride size
optional int32 blur_nrun = 106
Number of blur operation runs.
optional float markov_switch_prob = 107
Markov switching process probability.
optional bool mix_class_hierarchy = 108
Whether to mix class hierarchy and non class hierarchy.
optional bool use_same_family = 109
Whether to use the same family for different context.
optional bool shuffle_time = 110
Whether to shuffle temporal ordering.
optional bool use_new_class_hierarchy = 111
Whether to use new class hierarchy.
optional float prob_new = 200
New class parameter for constant prob sampler.
optional float crp_alpha = 201
Alpha parameter for CRP sampler.
optional float crp_theta = 202
Theta parameter for CRP sampler.
optional int32 distractor_nquery = 300
Number of queries for distractor images.
optional int32 distractor_nshot = 301
Number of shots for distractor images.
optional int32 distractor_nway = 302
Number of classes for distractor images.
optional bool random_box = 400
Add random box occluder.
optional string random_background = 401
Add random background. Choice of `none`, `uppsala`, `uppsala_double`.
optional bool random_background_random_apply = 402
Whether to randomly drop the background hue.
optional float random_background_apply_prob = 403
When random apply is true, use this probability to sample.
optional bool random_background_random_context = 404
When random apply is true, this will randomly choose a background and ignore the stage iD.
optional float random_background_gaussian_std = 405
Add gaussian noise to the random background.
message
experiment_config.proto:18optional string backbone_class = 1
CNN backbone.
optional string model_class = 2
Class name of model.
optional string dtype = 3
Data type.
optional int32 num_classes = 4
Total number of output classes.
optional int32 num_steps = 5
Number of training examples.
optional string memory_class = 6
Memory class.
optional bool freeze_backbone = 7
Whether or not to freeze backbone.
optional bool set_backbone_lr = 8
Whether to set a different learning rate for the backbone.
optional float backbone_lr_multiplier = 9
Set a learning rate multiplier for the backbone.
optional bool fix_unknown = 10
Whether to have a fixed unknown class ID.
optional float unknown_loss = 11
Multiplier for the unknown class loss.
optional bool renormalize_logits = 12
Whether or not to renormalize the logits.
optional int32 num_episodes = 13
Number of episodes for evaluation.
optional bool add_stage_id = 14
Whether to feed in stage ID.
optional bool in_stage = 15
Whether perform in stage training / testing.
optional bool roi_pooling_backbone = 16
Whether to perform ROI pooling with mask attention.
optional string sub_memory_class = 17
Sub memory class.
optional string sub_memory_class2 = 18
Sub memory class 2.
optional bool ssl_store_schedule = 19
Whether to store SSL with a schedule.
optional bool disable_loss_self_unk = 20
Disable binary xent for self-predicted unknowns.
optional string loss_fn = 21
"softmax" or "sigmoid".
optional c4_config = 101
4-layer CNN config.
optional resnet_config = 102
ResNet config.
optional protonet_config = 1000
ProtoNet config.
optional memory_net_config = 1001
Memory network config.
optional lstm_config = 1002
LSTM memory module config.
optional mann_config = 1003
Memory augmented module config.
optional oml_config = 1004
OML config.
optional hybrid_config = 1005
Hybrid model config.
optional optimizer_config = 2000
Optimizer config.
optional train_config = 3000
Training config.
message
hybrid_config.proto:7Used in:
optional bool use_pred_beta_gamma = 1
New ablation options:
optional bool use_feature_fuse = 2
optional bool use_feature_fuse_gate = 3
optional bool use_feature_scaling = 4
optional bool use_feature_memory_only = 5
optional bool use_element_fuse_gate = 6
optional string readout_type = 7
optional bool skip_unk_memory_update = 100
optional bool use_ssl = 101
optional bool use_ssl_beta_gamma_write = 102
optional bool use_ssl_temp = 103
message
lstm_config.proto:7Used in:
optional int32 hidden_dim = 1
optional int32 nstack = 2
optional bool layernorm = 3
message
mann_config.proto:7Used in:
optional int32 memory_dim = 1
optional int32 controller_dim = 2
optional int32 num_slots = 3
optional int32 num_reads = 4
optional int32 num_writes = 5
optional float memory_decay = 6
optional string controller_type = 7
optional int32 memory_attention_steps = 8
optional int32 num_write_steps = 9
optional int32 num_delay_steps = 10
optional string memory_attention_type = 11
optional int32 affinity_prop_steps = 12
optional string affinity_prop_similarity = 13
optional float affinity_prop_damping = 14
optional bool memory_layernorm = 15
optional bool controller_layernorm = 16
optional int32 controller_nstack = 17
optional string similarity_type = 18
optional int32 query_key_dim = 19
optional string readout_type = 20
optional bool use_unk_gate = 21
optional bool use_feat_gate = 22
message
memory_net_config.proto:7Used in:
optional int32 max_classes = 1
Maximum number of classes.
optional float radius_init = 2
Initialization for radius.
optional float alpha = 3
Not currently used.
optional int32 max_items = 4
For example based storage, maximum number of items.
optional float log_sigma_init = 5
Not currently used.
optional float log_lambda_init = 6
Not currently used.
optional int32 max_stages = 7
Not currently used.
optional int32 buffer_size = 8
Not currently used.
optional bool buffer_dropout = 9
Not currently used.
optional string similarity = 10
Similarity function.
optional float radius_init_write = 11
Initial value of the beta write.
optional string unknown_logits = 12
Choice of "radii" or "max"
optional float temp_init = 13
optional bool dense_update = 14
For GRU based models.
optional float gru_bias_init = 15
For GRU based models, initialize forget gate.
optional bool normalize_feature = 16
Whether to normalize feature prior to read/write.
optional bool use_ssl_beta_gamma_write = 17
Going to be coupled with hybrid_config.
optional bool fix_unknown = 18
Going to be coupled with main config.
optional int32 unknown_id = 19
ID for unknown.
message
oml_config.proto:7Used in:
repeated int32 num_filters = 1
Number of channels for each layer.
optional int32 num_classes = 2
Number of output classes.
optional float inner_lr = 3
Inner loop learning rate.
optional int32 inner_loop_truncate_steps = 4
Inner loop gradient truncation.
optional int32 repeat_steps = 5
Inner loop gradient truncation.
optional string inner_loop_loss = 6
Inner loop loss, "softmax" or "sigmoid" or "mix".
optional string unknown_logits = 7
How to compute unknown logits, "sum", "max", or "radii".
optional string unknown_output_type = 8
How to compute unknown logits, "softmax" or "sigmoid".
optional bool select_active_classes = 9
Select active classes.
optional bool learn_weight_init = 10
Learn initial weight.
optional bool cosine_classifier = 11
Use cosine.
optional bool classifier_bias = 12
Use bias.
optional bool semisupervised = 13
Whether to run semi-supervised.
message
optimizer_config.proto:7Used in:
optional string optimizer = 1
Optimizer name.
repeated float lr_list = 2
Learning rate decay values.
repeated int32 lr_decay_steps = 3
Learning rate decay steps.
optional int32 max_train_steps = 4
Total number of training steps.
optional int32 batch_size = 5
Training batch size.
optional int32 num_gpu = 6
Training number of GPUs.
optional float clip_norm = 7
optional int32 inner_loop_truncate_steps = 8
T-BPTT
optional bool lr_scaling = 9
Scale learning rate by the number of GPUs.
message
protonet_config.proto:7Used in:
optional string similarity = 1
Similarity function, euclidean or cosine.
optional bool cosine_a = 2
Whether to use cosine softmax in task A.
optional bool cosine_tau = 3
Whether to add a learnable scaling factor in task A.
optional bool cosine_softmax = 4
Whether to use cosine softmax in task B.
optional bool cosine_softmax_tau = 5
Whether to add a learnable scaling factor in task B
optional bool cosine_attention = 6
Whether to use cosine attention mechanism.
optional bool protos_phi = 7
Whether to gate the prototypes with learned parameters.
optional bool reinit_tau = 8
Whether to reinit tau variable for cosine A.
optional bool freeze_backbone = 9
Whether to freeze backbone network weights.
optional string scaling_module = 10
Which scaling module to use.
optional float static_scale = 11
Static scaling factor on the prototypes.
message
resnet_config.proto:7Used in:
optional int32 height = 1
Input image height.
optional int32 width = 2
Input image width.
optional int32 num_channels = 3
Input image number of channels.
repeated int32 num_residual_units = 4
Number of residual units.
repeated int32 num_filters = 5
Input image number of filters.
repeated int32 strides = 6
Stride for the initial convolution of each resolution stage.
optional int32 init_stride = 7
Initial convolution strides.
optional bool init_max_pool = 8
Whether doing max pooling in the initial convolution.
optional int32 init_filter = 9
Number of filters in the initial convolution.
optional bool use_bottleneck = 10
Whether to use bottleneck layer in each residual unit.
optional float weight_decay = 11
Weight decay.
optional string normalization = 12
Normalization scheme used in every layer.
optional int32 num_norm_groups = 13
Number of groups (used by GroupNorm).
optional bool global_avg_pool = 14
Whether perform global average pooling in the end.
optional string data_format = 15
Data format, NCHW or NHWC.
optional float leaky_relu = 16
Add leaky relu.
optional bool add_last_relu = 17
Add the final ReLU to the feature map.
optional bool add_dropout = 18
Dropout features.
optional float dropout_rate = 19
Dropout rate.
optional bool add_context_dropout = 20
Dropout features.
optional float context_dropout_rate = 21
Dropout rate.
optional float activation_scaling = 22
Activation scaling.
message
train_config.proto:7Used in:
optional int32 steps_per_val = 1
Number of steps for one validation run.
optional int32 steps_per_log = 2
Number of steps per output printing.
optional int32 steps_per_save = 3
Number of steps per mode checkpointing.
optional int32 pretrain_a_steps = 4
Number of steps to pretrain task A.