cityscapes数据集使用的一点点经验

做语义分割，刚入门，快被cityscapes数据集搞疯了。这里做个总结

数据集简介

Cityscapes数据集介绍

labels.py

eval_city:
图片地址列表：pred_fullpath_list
标签地址列表：gt_fullpath_list
[‘road’ ‘sidewalk’ ‘building’ ‘wall’ ‘fence’ ‘pole’ ‘light’ ‘sign’
‘vegetation’ ‘terrain’ ‘sky’ ‘person’ ‘rider’ ‘car’ ‘truck’ ‘bus’ ‘train’
‘motocycle’ ‘bicycle’]
pred path: /data/ysh/data/Cityscapes
pred path: /data/ysh/data/Cityscapes/leftImg8bit/val/frankfurt

The Cityscapes downloads already include such images
#   a) *color.png             : the class is encoded by its color
#   b) *labelIds.png          : the class is encoded by its ID
#   c) *instanceIds.png       : the class and the instance are encoded by an instance ID
# 
# With this tool, you can generate option
#   d) *labelTrainIds.png     : the class is encoded by its training ID
# This encoding might come handy for training purposes. You can use
# the file labes.py to define the training IDs that suit your needs.
# Note however, that once you submit or evaluate results, the regular
# IDs are needed.

frankfurt_000001_050149_leftImg8bit.png
~/code/MCD_DA/segmentation/test_output/gta-train2city-train_3ch—city-val/MCD-normal-drn_d_105-5.tar/label

val： leftImg8bit/val/frankfurt/frankfurt_000000_000294_leftImg8bit.png
…MCD-normal-drn_d_105-5.tar/label : frankfurt_000001_057954_leftImg8bit.png lindau_000032_000019_leftImg8bit.png
frankfurt/frankfurt_000001_007973_gtFine_labelIds.png

139
SAD
训练集
/data/ysh/code/SDA/filelist/cityscapes_imagelist_train.txt"
: bremen/bremen_000175_000019_leftImg8bit.png
/data/ysh/data/Cityscapes/leftImg8bit/train/