Linux
pytorch 0.4.1(python 2.7)
tensorboardX and tensorflow(just for visualization)
Your dataset should include thress sub-directory: lr* hr* img*
Warning: You should use the same format to name your images and PFM files
For training basic_res:
python main.py --maxdisp 144 --model basic_res --mask_disp 0.25 --datapath ./dataset -- epoches 50 --learning 1e-4 --savemodel ./trained_model --log_dir ./log
Decrease the mask_disp each time can improve the proformance. For example, the first time, mask_disp = 0.25, the second time, mask_disp = 0.15, the third time, mask_disp = 0
For training iternativenet:
The first stage:
python main.py --maxdisp 144 --model itnet --stage first --mask_disp 0.25 --datapath ./dataset -- epoches 50 --learning 1e-4 --savemodel ./trained_model --log_dir ./log
The second stage(distill stage):
python main.py --maxdisp 144 --model itnet --stage distill --loadmodel ./trained_model/checkpoint_50.tar --mask_disp 0 --datapath ./dataset -- epoches 50 --learning 1e-4 --savemodel ./trained_model --log_dir ./log
For testing
one image
python simple_test.py --model basic_res--img ./dataset/img_20190503_4/0m.png --lr ./dataset/lr_disp/0_disp.pfm --output_dir ./result --loadmodel ./trained_model/checkpoint_40.tar
or put your images and lr_pfms in your folders
python simple_test.py --model basic_res --img ./dataset/img_20190503_4 --lr ./dataset/lr_disp --output_dir ./result --loadmodel ./trained_model/checkpoint_40.tar
output: imagebasename_sr_vis.png, imagebasename_sr.pfm and imagebasename_res.pfm
If you use itnet, you should match the stage with the checkpoint. The output of the first stage is just like basic_res, and the output of the distill stage include imagebasename_res1.pfm imagebasename_res2.pfm in addition to basic output.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent