karolmajek/Axial-DeepLab-SWideRNet

1// Copyright 2021 The Deeplab2 Authors.2//3// Licensed under the Apache License, Version 2.0 (the "License");4// you may not use this file except in compliance with the License.5// You may obtain a copy of the License at6//7//     http://www.apache.org/licenses/LICENSE-2.08//9// Unless required by applicable law or agreed to in writing, software10// distributed under the License is distributed on an "AS IS" BASIS,11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.12// See the License for the specific language governing permissions and13// limitations under the License.14 15syntax = "proto2";16 17package deeplab2;18 19// Configure the dataset options.20message DatasetOptions {21  // Set the dataset. See dataset.py for supported datasets.22  optional string dataset = 1;23  // Set the dataset file pattern to be used with glob.24  repeated string file_pattern = 2;25  // Set the number of samples per batch. This must be a multiple of replicas.26  // E.g. batch_size = 8 on 4 GPUs equals a batch size of 2 on each GPU.27  optional int32 batch_size = 3 [default = 32];28  // Set the crop size as a list of [crop_height, crop_width].29  repeated int32 crop_size = 4;30  // Minimum value for resize. Can be 1) empty; or 2) an integer, indicating31  // the desired size of the shorter image side (either height or width); or32  // 3) a 2-tuple of (height, width), indicating the desired minimum value for33  // height and width after resize. Setting values to non-positive indicate34  // no minimum value would be used.35  repeated int32 min_resize_value = 5;36  // Maximum value for resize. Can be 1) empty; or 2) an integer, indicating37  // the maximum allowed size of the longer image side (either height or width);38  // or 3) a 2-tuple of (height, width), indicating the maximum allowed size39  // after resize. Setting values to non-positive indicates no maximum value40  // would be used.41  repeated int32 max_resize_value = 6;42  // Set the resizing factor.43  optional int32 resize_factor = 7;44 45  /* Augmentation options.*/46  message AugmentationOptions {47    // Set the minimum scale factor for augmentation. Default not to use.48    optional float min_scale_factor = 1 [default = 1.0];49    // Set the maximum scale factor for augmentation. Default not to use.50    optional float max_scale_factor = 2 [default = 1.0];51    // Set the scale factor step size for data augmentation.52    optional float scale_factor_step_size = 3 [default = 0.25];53    // The name of the AutoAugment policy to use.54    optional string autoaugment_policy_name = 4;55  }56  optional AugmentationOptions augmentations = 8;57  // Set the standard deviation used to generate Gaussian center ground-truth.58  optional float sigma = 9 [default = 8.0];59  // Set whether to use increased weights on small instances.60  optional bool increase_small_instance_weights = 10 [default = false];61  // Set the pixel threshold for small instances.62  optional int32 small_instance_threshold = 11 [default = 4096];63  // Set the small instance weight.64  optional float small_instance_weight = 12 [default = 3.0];65  // Set whether to use two frames togetehr (current frame + previous frame) as66  // input for video panoptic segmentation.67  optional bool use_two_frames = 13 [default = false];68  // Whether to decode the groundtruth label. Some dataset splits (e.g., test69  // set) may not contain any groundtruth label. In that case, set this field70  // to false to avoid decoding non-existing groundtruth label.71  optional bool decode_groundtruth_label = 14 [default = true];72  // Whether the model needs thing_id_mask annotations. When True, we will73  // additionally return mask annotation for each `thing` instance, encoded with74  // a unique thing_id. This ground-truth annotation could be used to learn a75  // better segmentation mask for each instance. `thing_id` indicates the number76  // of unique thing-ID to each instance in an image, starting the counting from77  // 0 (default: False).78  optional bool thing_id_mask_annotations = 15 [default = false];79  // Set the maximum number of possible thing instances per image. It is used80  // together when enabling generation of thing_id_mask_annotations (= True),81  // representing the maximum thing ID encoded in the thing_id_mask.82  optional int32 max_thing_id = 16 [default = 128];83  // Set whether to use the next frame together with the current frame for video84  // panoptic segmentation (VPS). This field also controls using two-frame as85  // input for VPS. Note that `use_two_frames` is adopted in Motion-DeepLab,86  // while `use_next_frame` is used in ViP-DeepLab.87  optional bool use_next_frame = 17 [default = false];88}89