使用dilated conv和resnet组合的语义分割网络，性能优异
# 165次迭代后的CamVid校准数据集上的精度
('FreqW Acc : \t', 0.78070305048730892)
# 测试集中0.59891324612048458
('Overall Acc: \t', 0.85877188760542722)
# 测试集中0.71824451597520267
('Mean Acc : \t', 0.61283971438429574)
# 测试集中0.44491145002537369
('Mean IoU : \t', 0.50848648789574569)
# 测试集中0.32908427092179399
(0, 0.73215407785969255)
# Sky天空 测试集中0.63814217101601467
(1, 0.71535658519148426)
# Building建筑物 测试集中0.56336332903217423
(2, 0.053113375831408939)
# Pole路灯 测试集中0.055465076303629596
(3, 0.95854814284974066)
# Road_marking 道路标记 测试集中0.86068851093222054
(4, 0.82224281506323227)
# Road道路 测试集中0.60687183556006807
(5, 0.84803394728268744)
# Pavement人行道 测试集中0.43652442899653782
(6, 0.15571288956455642)
# Tree树木 测试集中0.046014547880138687
(7, 0.4141385732203261)
# SignSymbol 交通信号 测试集中0.091659446125291782
(8, 0.61702347918649891)
# Fence 测试集中0.3770718777383672
(9, 0.12849316561387919)
# Car汽车 测试集中0.088796649002151298
(10, 0.54299787839035307)
# Pedestrian 行人 测试集中0.025057527717900444
(11, 0.11402292469508754)
# Bicyclist 自行车手 测试集中0.15935585075703296
(12, nan)