config.py

import albumentations as A
import cv2
import torch

from albumentations.pytorch import ToTensorV2
from utils import seed_everything

DATASET = 'PASCAL_VOC'
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
# seed_everything()  # If you want deterministic behavior
NUM_WORKERS = 0
BATCH_SIZE = 2
DIV = 32
IMAGE_SIZES = [416, 416, 416, 608, 608]
S = [[x//DIV, x//DIV*2, x//DIV*4] for x in IMAGE_SIZES]
NUM_CLASSES = 20
LEARNING_RATE = 1e-5
WEIGHT_DECAY = 1e-4
NUM_EPOCHS = 10
CONF_THRESHOLD = 0.05
MAP_IOU_THRESH = 0.5
NMS_IOU_THRESH = 0.45
PIN_MEMORY = True
LOAD_MODEL = False
SAVE_MODEL = True
CHECKPOINT_FILE = "checkpoint.pth.tar"
IMG_DIR = DATASET + "/images/"
LABEL_DIR = DATASET + "/labels/"
MOSAIC_PROB = 0.75
INFERENCE_IMAGE_SIZE = 416

ANCHORS = [
    [(0.28, 0.22), (0.38, 0.48), (0.9, 0.78)],
    [(0.07, 0.15), (0.15, 0.11), (0.14, 0.29)],
    [(0.02, 0.03), (0.04, 0.07), (0.08, 0.06)],
]  # Note these have been rescaled to be between [0, 1]

means = [0.45484068, 0.43406072, 0.40103856]
stds = [0.23936155, 0.23471538, 0.23876129]

scale = 1.1
def train_transform(IMAGE_SIZE):
    train_transforms = A.Compose(
        [
            A.LongestMaxSize(max_size=int(IMAGE_SIZE * scale)),
            A.PadIfNeeded(
                min_height=int(IMAGE_SIZE * scale),
                min_width=int(IMAGE_SIZE * scale),
                border_mode=cv2.BORDER_CONSTANT,
            ),
            A.Rotate(limit = 10, interpolation=1, border_mode=4),
            A.RandomCrop(width=IMAGE_SIZE, height=IMAGE_SIZE),
            A.ColorJitter(brightness=0.6, contrast=0.6, saturation=0.6, hue=0.6, p=0.4),
            A.OneOf(
                [
                    A.ShiftScaleRotate(
                        rotate_limit=20, p=0.5, border_mode=cv2.BORDER_CONSTANT
                    ),
                    # A.Affine(shear=15, p=0.5, mode="constant"),
                ],
                p=1.0,
            ),
            A.HorizontalFlip(p=0.5),
            A.Blur(p=0.1),
            A.CLAHE(p=0.1),
            A.Posterize(p=0.1),
            A.ToGray(p=0.1),
            A.ChannelShuffle(p=0.05),
            A.Normalize(mean=means, std=stds, max_pixel_value=255,),
            ToTensorV2(),
        ],
        bbox_params=A.BboxParams(format="yolo", min_visibility=0.4, label_fields=[],),
    )

    return(train_transforms)

def test_transform(IMAGE_SIZE=416):
    test_transforms = A.Compose(
        [
            A.LongestMaxSize(max_size=IMAGE_SIZE),
            A.PadIfNeeded(
                min_height=IMAGE_SIZE, min_width=IMAGE_SIZE, border_mode=cv2.BORDER_CONSTANT
            ),
            A.Normalize(mean=means, std=stds, max_pixel_value=255,),
            ToTensorV2(),
        ],
        bbox_params=A.BboxParams(format="yolo", min_visibility=0.4, label_fields=[]),
    )
    return(test_transforms)


PASCAL_CLASSES = [
    "aeroplane",
    "bicycle",
    "bird",
    "boat",
    "bottle",
    "bus",
    "car",
    "cat",
    "chair",
    "cow",
    "diningtable",
    "dog",
    "horse",
    "motorbike",
    "person",
    "pottedplant",
    "sheep",
    "sofa",
    "train",
    "tvmonitor"
]