Super-resolution based on EDSR
python 3.5
- Numpy >= 1.14.0
- Tensorflow >= 1.3.0
- horovod
- Data Pipeline
- TFrecords
build_tfrecords.py
tfrecords_reader.py
build_train_tfrecords.sh
data_generator.py
- Training Pipeline
multi_trainer.py
config_training.py
- Testing Pipeline
tester.py
config_testing.py
- EDSR
models/EDSR.py
models/ops.py
- Utils
config_helpers.py
- Copy or build TFrecords for training data
- Copy ready-made TFrecords for NTIRE2017 and NTIRE2018
- Or, make TFrecords from
.png
files (Seebuild_train_tfrecords.sh
)
- Train the model
Modify config_training.py
as you want.
Then, run *_trainer*.py
- Test the model
Copy .png
files from nas/dataset/NTIRE*/
.
Modify config_testing.py
as you want.
Then, run *_tester.py
CUDA_VISIBLE_DEVICES=0,1,2,3 \
mpirun -np 4 \
-H localhost:4 \
-bind-to none -map-by slot \
-x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH \
python3 multi_trainer.py