Skip to content

Configuration Variables

Jump to bottom
jkoessle edited this page Aug 29, 2023 · 3 revisions

Here all configuration variables from the object detection config.py are explained in more detail.

  • DEBUG := (bool) specifies debugging state; defaults to True.

  • OBJECT_DETECTION := (bool) specifies object detection state; defaults to True.

  • ANNOTATIONS_ONLY := (bool) specifies whether only annotations should be created during preprocessing and skips the actual preprocessing; defaults to False.

  • AUTOMATE_TFR_SCRIPT := (bool) specifies whether to automate TFR script, needs path of TENSORFLOW_MODELS_DIR and calls coco script; defaults to True.

  • VDD_PREPROCESSING := (bool) specifies which encoding to use; defaults to False.

  • KEEP_AXIS := (bool) specifies whether axis should be kept for VDD encoding; defaults to False.

  • WINDOWS_SYSTEM := (bool) specifies whether system is Windows or unix-based; defaults to True.

  • MINE_CONSTRAINTS := (bool) specifies whether to mine constraints with MINERful; defaults to True.

  • CONSTRAINTS_DIR := (str) specifies directory of already mined constraints if MINE_CONSTRAINTS is False.

  • N_WINDOWS := (int) specifies number of windows for WINSIM encoding; defaults to 200.

  • DEFAULT_DATA_DIR := (str) specifies default data output directory.

  • DEFAULT_LOG_DIR := (str) specifies event log directory.

  • TFR_RECORDS_DIR := (str) specifies directory where to save TFR files.

  • TENSORFLOW_MODELS_DIR := (str) specifies TensorFlow model garden directory. The TF model garden is available here.

  • MINERFUL_SCRIPTS_DIR := (str) specifies MINERful directory. MINERful is available here

  • OUTPUT_PREFIX := (str) specifies output prefix for TFR file.

  • DRIFT_TYPES := (list[str]) specifies drift types; defaults to ["sudden", "gradual", "incremental", "recurring"].

  • DISTANCE_MEASURE := (str) specifies similarity measure; defaults to "cos". Can be one of ["fro","nuc","inf","l2","cos","earth"].

  • COLOR := (str) specifies whether images should be in color or grayscale; defaults to "color".

  • RESIZE_SUDDEN_BBOX := (bool) specifies whether to resize bounding box for sudden drifts; defaults to True.

  • RESIZE_VALUE := (int) specifies resize value for resizing bboxes for sudden drifts; defaults to 5.

  • SUB_L := (int) specifies number of sublogs for VDD encoding; defaults to 100.

  • SLI_BY := (int) specifies sliding by value for VDD encoding; defaults to 50.

  • CP_ALL := (bool) specifies whether to detect all changepoints for VDD encoding; defaults to True.

  • FACTOR := (int) specifies training length factor; defaults to 500.

  • TRAIN_EXAMPLES := (int) specifies number of training examples.

  • EVAL_EXAMPLES := (int) specifies number of validation examples.

  • TRAIN_BATCH_SIZE := (int) specifies training batch size; defaults to 64.

  • EVAL_BATCH_SIZE := (int) specifies validation batch size; defaults to 32.

  • STEPS_PER_LOOP := (int) specifies training steps per iteration; defaults to TRAIN_EXAMPLES // TRAIN_BATCH_SIZE.

  • TRAIN_STEPS := (int) specifies total training steps; defaults to FACTOR * STEPS_PER_LOOP.

  • VAL_STEPS := (int) specifies validation steps per iteration; defaults to EVAL_EXAMPLES // EVAL_BATCH_SIZE.

  • SUMMARY_INTERVAL := (int) specifies summary interval; defaults to STEPS_PER_LOOP.

  • CP_INTERVAL := (int) specifies checkpoint interval; defaults to STEPS_PER_LOOP.

  • VAL_INTERVAL := (int) specifies validation interval; defaults to STEPS_PER_LOOP.

  • EVAL_THRESHOLD := (float) specifies prediction confidence threshold; defaults to 0.5.

  • IMAGE_SIZE := (tuple[int]) specifies image size for training; defaults to (256, 256).

  • TARGETSIZE := (int) specifies targetsize for training images; defaults to 256.

  • N_CLASSES := (int) specifies number of classes; defaults to len(DRIFT_TYPES).

  • SCALE_MAX := (float) specifies scale augmentation maximum; defaults to 2.0.

  • SCALE_MIN := (float) specifies scale augmentation minimum; defaults to 0.1.

  • LR_DECAY := (bool) specifies whether learning rate should decay during training; defaults to True.

  • LR_INITIAL := (float) specifies initial learning rate; defaults to 1e-3.

  • LR_WARMUP := (float) specifies warmup learning rate; defaults to 2.5e-4.

  • LR_WARMUP_STEPS := (int) specifies number of learning rate warmup steps; defaults to 0.1 * TRAIN_STEPS.

  • BEST_CP_METRIC := (str) specifies best changepoint metric; defaults to "AP".

  • BEST_CP_METRIC_COMP := (str) specifies comparison for cp metric; defaults to "higher".

  • OPTIMIZER_TYPE := (str) specifies optimizer; defaults to "sgd".

  • LR_TYPE := (str) specifies learning rate decay type; defaults to "stepwise".

  • SGD_MOMENTUM := (float) specifies momentum for SGD; defaults to 0.9.

  • SGD_CLIPNORM := (float) specifies clipnorm for SGD; defaults to 10.0.

  • ADAM_BETA_1 := (float) specifies beta_1 for adam; defaults to 0.9.

  • ADAM_BETA_2 := (float) specifies beta_2 for adam; defaults to 0.999.

  • STEPWISE_BOUNDARIES := (list[float]) specifies stepwise boundaries for decay; defaults to [0.95 * TRAIN_STEPS,0.98 * TRAIN_STEPS].

  • STEPWISE_VALUES := (list[float]) specifies stepwise values; defaults to [0.32 * TRAIN_BATCH_SIZE / 256.0, 0.032 * TRAIN_BATCH_SIZE / 256.0, 0.0032 * TRAIN_BATCH_SIZE / 256.0].

  • MODEL_SELECTION := (str) specifies model; defaults to "retinanet_spinenet_coco".

  • SPINENET_ID := (int) specifies which spine net to use; defaults to "143".

  • TRAIN_DATA_DIR := (str) specifies training data file in TFR format.

  • EVAL_DATA_DIR := (str) specifies validation data file in TFR format.

  • MODEL_PATH := (str) specifies logging directory.

  • DEFAULT_OUTPUT_DIR := (str) specifies default output directory for results.

  • TRAINED_MODEL_PATH := (str) specifies path of pretrained/trained model.

  • TEST_IMAGE_DATA_DIR := (str) specifies directory where evaluation images are saved.

  • RELATIVE_LAG := (list[float]) specifies relative lag values for evaluation; defaults to [0.01, 0.025, 0.05, 0.1, 0.15, 0.2].

  • EVAL_MODE := (str) specifies evaluation mode, for naming purposes.

  • PRODRIFT_DIR := (str) specifies directory where ProDrift is stored.

  • VDD_DIR := (str) specifies directory where VDD is stored.

Clone this wiki locally