synChange Dataset Generation

Installation

0. Prerequisites

• Python 3.7 or later
• Pytorch 1.8.1 or later
• cuda 11.1 or later

1. Install requirements

pip install -r requirements.txt

2. Install cupy

pip install cupy-cuda111 --no-cache-dir

Prepare Pre-trained Model

cd GLU-ChangeNet-Pytorch
sudo scp -r [email protected]:/home/rit/E2EChangeDet/GLU-ChangeNet-Pytorch/pre_trained_models .

Prepare Train set

python save_change_training_dataset_to_disk.py --save_dir result --plot True

or just download one

cd ..
mkdir dataset
cd dataset
mkdir dataset/train_datasets
cd train_datasets
sudo scp -r [email protected]:/home/rit/E2EChangeDet/dataset/train_datasets/synthetic .
cd synthetic/flow
ln -s static new
ln -s ststic replaced
ln -s static missing

Prepare Test set

cd ../../.. 
# In 'dataset' folder
sudo scp -r [email protected]:/home/rit/E2EChangeDet/dataset/test_datasets .

Init Training for Single-class Change Detection

cd ../GLU-ChangeNet-Pytorch
python train_GLUChangeNet.py \
--training_data_dir ../dataset/train_datasets/synthetic_dataset \
--evaluation_data_dir ../dataset/test_datasets \
--pretrained pre_trained_models/GLUNet_DPED_CityScape_ADE.pth \
--n_threads 4 --plot_interval 1 --split_ratio 0.90 --split2_ratio 0.5 \
--trainset_list synthetic vl_cmu_cd \
--testset_list vl_cmu_cd \
--lr 0.0002 --n_epoch 25 \
--name_exp joint_synthetic_vl-cmu-cd

Init Training for Multi-class Change Detection

cd ../GLU-ChangeNet-Pytorch
--training_data_dir ../dataset/train_datasets/synthetic_dataset \
--evaluation_data_dir ../dataset/test_datasets \
--pretrained pre_trained_models/GLUNet_DPED_CityScape_ADE.pth \
--n_threads 4 --plot_interval 1 --split_ratio 0.90 --split2_ratio 0.5 \
--trainset_list synthetic \
--testset_list changesim \
--lr 0.0002 --n_epoch 25 \
--name_exp train_synthetic_test_changesim_multiclass \
--multi_class

SynChange Dataset Generation

python save_change_training_dataset_to_disk.py