The conigs of the state-of-the-art methods¶

In “/configs/”, we write four file to save the configs of the state-of-the-art methods. And the file type is .yml The configs combine experiment name, weight, logger, dataset, model, transforms, optimizer, and Validation details.

Take the “baseline_resnet50” for example.

Experiment name

EXP_NAME: the name of this method.
```
EXP_NAME: "Baseline_ResNet50"
```
Resume weight

RESUME_WEIGHT is set ~
```
RESUME_WEIGHT: ~
```
Logger

LOGGER: the logger of this process.
```
LOGGER: 
  NAME: "txt_logger"
```
Dataset

DATASET: the dataset
- NAME: the name of the dataset
- ROOT: the root of the dataset, you should change it to your own root in advance
- TRAIN: the parameters about the processing of training
  - BATCH_SIZE: the batch size
  - POSITIVE: the positive
  - PIN_MEMORY: is bool, True or False
  - SHUFFLE: is bool, True or False
  - NUM_WORKERS: the number of workers
- TEST: the parameters about the processing of testing
  - BATCH_SIZE: the batch size
  - POSITIVE: the positive
  - PIN_MEMORY: is bool, True or False
  - SHUFFLE: is bool, True or False
  - NUM_WORKERS: the number of workers

DATASET:
  NAME: "CUB_200_2011"
  ROOT: "/data/wangxinran/dataset/"
  TRAIN: 
    BATCH_SIZE: 2
    POSITIVE: 0
    PIN_MEMORY: True
    SHUFFLE: True
    NUM_WORKERS: 4
  TEST: 
    BATCH_SIZE: 2
    POSITIVE: 0
    PIN_MEMORY: False
    SHUFFLE: False
    NUM_WORKERS: 4

Model

MODEL: the model
- NAME: the name of the model
- CLASS_NUM: the number of the class
- CRITERIONS: the loss
  - name: the name of loss function
  - args: the args are determined by loss
  - w: the weight
- BACKBONE: the backbone of the model, the parameters about the args are set according the corresonding file.
  - NAME: the name of the bockbone
  - ARGS:
    - pretrained: is bool, True or False
    - del_keys: is a list of the del_keys
- ENCODING: the encoding of the model, the parameters about the args are set according the corresonding file.
  - NAME: the name of the encoding
- NECKS: the neck of the model
  - NAME: the name of the neck
- HEADS: the head of the model, the parameters about the ARGS are set according the corresonding file.
  - NAME: the name of the head
  - ARGS: the args are determined by the classifier.

MODEL:
  NAME: "ResNet50"
  CLASS_NUM: 200
  CRITERIONS: 
    - name: "cross_entropy_loss"
      args: []
      w: 1.0
  BACKBONE:
    NAME: "resnet50"
    ARGS:
      - pretrained: True
      - del_keys: []
  ENCODING:
    NAME: "global_avg_pooling"
  NECKS:
    NAME: ~
  HEADS:
    NAME: "classifier_1fc"
    ARGS:
      - in_dim: 
        - 2048

Transforms

TRANSFORMS: the transforms
- TRAIN: the parameters about the process of training
  - name: the name of the transform type
  - other parameters: according the transform type to set the corresonding parameters.
- TEST: the parameters about the process of testing
  - name: the name of the transform type
  - other parameters: according the transform type to set the corresonding parameters.

TRANSFORMS: 
  TRAIN: 
    - name: "resize"
      size: 
        - 600
        - 600
    - name: "random_crop"
      size: 448
      padding: 8
    - name: "random_horizontal_flip"
      prob: 0.5
    - name: "to_tensor"
    - name: "normalize"
      mean: 
         - 0.5
         - 0.5
         - 0.5
      std: 
        - 0.5
        - 0.5
        - 0.5
  TEST:
    - name: "resize"
      size: 
        - 600
        - 600
    - name: "center_crop"
      size: 448
    - name: "to_tensor"
    - name: "normalize"
      mean: 
         - 0.5
         - 0.5
         - 0.5
      std: 
        - 0.5
        - 0.5
        - 0.5

Optimizer

OPTIMIZER: the optimizer
- NAME: the name of the optimizer
- MOMENTUM: the momentum of the optimizer
- LR: the learning rate
  - backbone: the learning rate of the backbone
  - encoding: the learning rate of the encoding
  - necks: the learning rate of the neck
  - heads: the learning rate of the head

OPTIMIZER:
  NAME: "SGD"
  MOMENTUM: 0.9
  LR: 
    backbone: 0.0002
    encoding: 0.002
    necks: 0.002
    heads: 0.002

Iteration number

ITERATION_NUM: the Iteration number

ITERATION_NUM: ~

Epoch number

EPOCH_NUM: the epcoh number

START_EPOCH: 0

Update strategy

UPDATE_STRATEGY: the update strategy

UPDATE_STRATEGY: "general_updating"

Per iteration

PER_ITERATION: the per iteration

PER_ITERATION: ~

Per epoch

PER_EPOCH： the per per iteration

PER_EPOCH: ~

Metrics

METRICS: the metrics
- name: the name of the metric type
- other parameters: according the metrics type to set the corresonding parameters.

METRICS: 
  - name: "accuracy(topk=1)"
    metric: "accuracy"
    top_k: 1
    threshold: ~
  - name: "accuracy(topk=5)"
    metric: "accuracy"
    top_k: 5
    threshold: ~
  - name: "recall(threshold=0.5)"
    metric: "recall"
    top_k: ~
    threshold: 0.5
  - name: "precision(threshold=0.5)"
    metric: "precision"
    top_k: ~
    threshold: 0.5

Interpreter

INTERPRETER: the interpreter
- NAME: the name of the interpreter
- METHOD: the method of the interpreter
- TARGET_LAYERS: the target layers