task: --> tianchi
- image segmentation
- salient object detection
- use ResNeXT as encoder;
- use UNet framework;
- use DAHead as decode_head;
see seg model
mIoU:
use make_dataset.py
to make the .tfrecord
files
python -W ignore train.py --batch_size $batch_size --niter $niter --lr $lr
- refine the u2net
- add: EGNet
- data aug
- model fusion