cuda_available = torch.cuda.is_available() BATCH_SIZE = 32 MAX_EPOCHS = 800 INIT_LR = 0.001 WEIGHT_DECAY = 0.00005 LR_DROP_MILESTONES = [400, 600] train_file_dir = 'PATH/TO/TRAINING/ANNOTATIONS' valid_file_dir = 'PATH/TO/VALIDATION/ANNOTATIONS' train_img_dir = 'PATH/TO/876x657/IMAGES' valid_img_dir = 'PATH/TO/768x576/IMAGES' save_path = 'PATH/TO/SAVE/TRAINING/MODELS' train_dataset = TrafficLightDataset(csv_file=train_file_dir, img_dir=train_img_dir) valid_dataset = TrafficLightDataset(csv_file=valid_file_dir, img_dir=valid_img_dir) train_dataloader = DataLoader(train_dataset, batch_size=BATCH_SIZE, shuffle=True, num_workers=8) valid_dataloader = DataLoader(valid_dataset, batch_size=1, shuffle=False, num_workers=2) net = LYTNet() if cuda_available:
import time from torch.utils.data import DataLoader from LYTNet import LYTNet from loss import my_loss from helpers import direction_performance, display_image from dataset import TrafficLightDataset import matplotlib.pyplot as plt import numpy as np cuda_available = torch.cuda.is_available() test_file_loc = '/home/user/Desktop/eyeDoPy/Annotations/testing_file.csv' test_image_directory = '/home/user/Desktop/eyeDoPy/PTL_Dataset_768x576/' MODEL_PATH = '/home/user/Desktop/eyeDoPy/Model/_final_weights' dataset = TrafficLightDataset(csv_file=test_file_loc, img_dir=test_image_directory) dataloader = DataLoader(dataset, batch_size=1, shuffle=False, num_workers=2) #load weights net = LYTNet() checkpoint = torch.load(MODEL_PATH) net.load_state_dict(checkpoint) net.eval() if cuda_available: net = net.cuda() loss_fn = my_loss #storing data running_loss = 0
valid_file_root = '/home/mv01/Desktop/ISEF 2018/new 5-fold files/valid_file_' image_directory = '/home/mv01/Desktop/ISEF 2018/resized_photos_512_384' MODEL_SAVE_PATH = '/home/mv01/Desktop/ISEF 2018/train_cycle_14' WEIGHT_LOAD_PATH = '/home/mv01/Desktop/ISEF 2018/train_cycle_14_epoch_50_weights2' #these save the data for each of the 10 folds fold_valid_accuracies = [] fold_valid_losses = [] fold_valid_angle = [] fold_valid_start = [] fold_valid_end = [] #10-fold cross validation for i in range(3,4): train_file_loc = train_file_root + str(i+1) + '.csv' train_dataset = TrafficLightDataset(csv_file = train_file_loc, root_dir = image_directory) valid_file_loc = valid_file_root + str(i+1) + '.csv' valid_dataset = ValidDataset(csv_file = valid_file_loc, root_dir = image_directory) train_dataloader = DataLoader(train_dataset, batch_size=BATCH_SIZE, shuffle=True, num_workers=8) valid_dataloader = DataLoader(valid_dataset, batch_size=1, shuffle=False, num_workers=2) net = MyNet() #checkpoint = torch.load(WEIGHT_LOAD_PATH) #net.load_state_dict(checkpoint['state_dict']) if cuda_available: net = net.cuda() loss_fn = my_loss