Startup.py

import os

import torch
from PIL import Image, ImageFile

device = torch.device("cuda:0")
ImageFile.LOAD_TRUNCATED_IMAGES = True


#INPUT_ROOT = r'input'
INPUT_ROOT = r'E:\Temp\Datasets\APTOS 2019\input'

IMAGE_FOLDER = r'train_images_t3_512'
#IMAGE_FOLDER = r'test15_t3_512'
#IMAGE_FOLDER = r'train15_t3_512'
#IMAGE_FOLDER = r'traintest15_t3_512'

TRAINING_CSV = 'trainLabels19.csv'
#TRAINING_CSV = 'testLabels15.csv'
#TRAINING_CSV = 'trainLabels15.csv'
#TRAINING_CSV = 'traintest15.csv'

IMAGE_ID = 'id_code'# 'image' # id_code
#IMAGE_ID = 'image'

IMAGE_LABEL = 'diagnosis'# 'level' # diagnosis
#IMAGE_LABEL = 'level'

MODEL_ROOT = r'models'
MODEL_NAME = r'efficientnet_15_e19.pth'
MODEL_PATH = os.path.join(MODEL_ROOT, MODEL_NAME)
STATE_PATH = os.path.join(MODEL_ROOT, 'training_' + MODEL_NAME)

NEW_MODEL = False
NEW_TRAINING = False

VALIDATION_PERCENTAGE = 0.1

# TODO figure out if large images work
IMG_SIZE = 224
#IMG_SIZE = 512

NUM_CLASSES = 5

BATCH_SIZE = 8
STEP_FREQ = 2
EPOCHS = 200
NUM_WORKERS = 8

LEARNING_RATE = 1e-4

L2_LOSS = 4e-6