Skip to content

Latest commit

 

History

History

training_code

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
README.md
README.md
 
 
convert_to_tf_record.py
convert_to_tf_record.py
 
 
crf.txt
crf.txt
 
 
crf_net_v2.npy
crf_net_v2.npy
 
 
dataset.py
dataset.py
 
 
dequantization_net.py
dequantization_net.py
 
 
dorfCurves.txt
dorfCurves.txt
 
 
emor.txt
emor.txt
 
 
finetune_real_dataset.py
finetune_real_dataset.py
 
 
hallucination_net.py
hallucination_net.py
 
 
i_dataset_test.pkl
i_dataset_test.pkl
 
 
i_dataset_train.pkl
i_dataset_train.pkl
 
 
invemor.txt
invemor.txt
 
 
linearization_net.py
linearization_net.py
 
 
net.py
net.py
 
 
refinement_net.py
refinement_net.py
 
 
train_dequantization_net.py
train_dequantization_net.py
 
 
train_hallucination_net.py
train_hallucination_net.py
 
 
train_linearization_net.py
train_linearization_net.py
 
 
util.py
util.py
 
 

README.md

Training

  1. Download the pre-trained weights of vgg16 and vgg16_places365_weights
  2. Download the training data of HDR-Synth and HDR-Real

Training of the Dequantization-Net

python train_dequantization_net.py --logdir_path [output_deq_ckpt_path] --hdr_prefix [hdr_synth_training_data_path]

Training of the Linearization-Net

python train_linearization_net.py --logdir_path [output_lin_ckpt_path] --hdr_prefix [hdr_synth_training_data_path]

Training of the Hallucination-Net

python train_hallucination_net.py --logdir_path [output_hal_ckpt_path] --hdr_prefix [hdr_synth_training_data_path]

Joint training of the entire pipeline and Refinement-Net

  1. Convert the real HDR-jpg paired data into tfrecords for training.
python convert_to_tf_record.py
  1. Joint training of the entire pipeline and Refinement-Net
python finetune_real_dataset.py --logdir_path [output_hal_ckpt_path] --tfrecords_path [converted_tfrecords_path] --deq_ckpt [pretrained_deq_ckpt] --lin_ckpt [pretrained_lin_ckpt] --deq_ckpt [pretrained_hal_ckpt]