configs.py

YOLO_TYPE                   = "yolov4" # yolov4 or yolov3
YOLO_FRAMEWORK              = "tf" # "tf" or "trt"
YOLO_V3_WEIGHTS             = "model_data/yolov3.weights"
YOLO_V4_WEIGHTS             = "model_data/yolov4.weights"
YOLO_V3_TINY_WEIGHTS        = "model_data/yolov3-tiny.weights"
YOLO_V4_TINY_WEIGHTS        = "model_data/yolov4-tiny.weights"
YOLO_TRT_QUANTIZE_MODE      = "INT8" # INT8, FP16, FP32
YOLO_CUSTOM_WEIGHTS         = True # "checkpoints/yolov3_custom" # used in evaluate_mAP.py and custom model detection, if not using leave False
YOLO_COCO_CLASSES           = "model_data/coco/coco.names"
YOLO_STRIDES                = [8, 16, 32]
YOLO_IOU_LOSS_THRESH        = 0.6
YOLO_ANCHOR_PER_SCALE       = 3
YOLO_MAX_BBOX_PER_SCALE     = 100
YOLO_INPUT_SIZE             = 832
YOLO_ANCHORS                = [[[12,  16], [19,   36], [40,   28]],
                               [[36,  75], [76,   55], [72,  146]],
                               [[142, 110], [192, 243], [459, 401]]]

# Train options
# TRAIN_YOLO_TINY             = False
# TRAIN_SAVE_BEST_ONLY        = True # saves only best model according validation loss (True recommended)
# TRAIN_SAVE_CHECKPOINT       = False # saves all best validated checkpoints in training process (may require a lot disk space) (False recommended)
TRAIN_CLASSES               = "classes.txt"
# TRAIN_ANNOT_PATH            = "model_data/kaggle/traffic_drum_train.txt"
TRAIN_LOGDIR                = "log"
TRAIN_CHECKPOINTS_FOLDER    = "/home/aaryaman_bhardwaj/Documents/ML/Yolo-v3/checkpoints"
TRAIN_MODEL_NAME            = f"{YOLO_TYPE}_cone_1"
WEIGHTS_FILE                = "/home/aaryaman_bhardwaj/Documents/ML/Yolo-v3/checkpoints/yolov4_cone_1"
# TRAIN_LOAD_IMAGES_TO_RAM    = True # With True faster training, but need more RAM
# TRAIN_BATCH_SIZE            = 2
# TRAIN_INPUT_SIZE            = 832
# TRAIN_DATA_AUG              = True
# TRAIN_TRANSFER              = False
# TRAIN_FROM_CHECKPOINT       = True # "checkpoints/yolov3_custom"
# TRAIN_LR_INIT               = 1e-4
# TRAIN_LR_END                = 1e-6
# TRAIN_WARMUP_EPOCHS         = 2
# TRAIN_EPOCHS                = 2

# TEST options
# TEST_ANNOT_PATH             = "model_data/kaggle/traffic_drum_test.txt"
# TEST_BATCH_SIZE             = TRAIN_BATCH_SIZE  #2
# TEST_INPUT_SIZE             = TRAIN_INPUT_SIZE  #832
# TEST_DATA_AUG               = True
TEST_DECTECTED_IMAGE_PATH   = "IMAGES/temp_test_detected"
# TEST_SCORE_THRESHOLD        = 0.55
# TEST_IOU_THRESHOLD          = 0.6