task: --> tianchi

1. image segmentation
2. salient object detection

• use ResNeXT | Res2Net | SeResNeXT as encoder;
• use UNet | R2UNet framework;
• use DAHead | PSPHead as decode_head;

see seg model

• use U2Net | EGNet | CSF+Res2Net;

see det model

use make_dataset.py to make the .tfrecord files

python -W ignore train.py --batch_size $batch_size --niter $niter --lr $lr

• data aug
• model fusion