About

A tensorflow implementation based on SegNet with a little change trained on CamVid dataset.

The CamVid dataset can be as follows:

|--CamVid
     |--train       (367)
          |--*.png
     |--trainannot  (367)
          |--*.png
     |--val         (101)
          |--*.png  
     |--valannot    (101)
          |--*.png
     |--test        (233)
          |--*.png
     |--testannot   (233)
          |--*.png

To tfrecord

python to_tfrecord.py

Train

python train.py

Test

python predict.py

Results(trained on train set)

Before optimization

• prediction on train set

  

• prediction on test set

  

• train loss and mIoU