🎉🎉🎉 Welcome to the SDDR GitHub repository! 🎉🎉🎉

Authors: Jiaqi Li¹, Yiran Wang¹, Jinghong Zheng¹, Zihao Huang¹, Ke Xian¹, Zhiguo Cao¹, Jianming Zhang²,

Institutes: ¹Huazhong University of Science and Technology, ²Adobe Research

pip install -r requirements.txt

As an example of reasoning on middlebury, the input parameters are the weight path, the rgb folder path, the output path, and the folder path predicted by the base high and low resolutions.

python demo.py  --weight checkpoints/model_dict_1_5600.pt \
		--rgb demo_input/Middlebury2021/rgb \
		--output demo_output/Middlebury2021/LeRes_Fusion \
		--low demo_input/Middlebury2021/LeRes/low_448 \
		--high demo_input/Middlebury2021/LeRes/high_1920 

# or you can simply run this for middlebury2021 inference(all parameters are set to default)
python demo.py

The official metrics can be verified directly using the previous_evaluate_mid21.py (the middlebury2021 dataset needs to be unzipped to demo_input/Middlebury2021/GTData)

The two-stage prediction will be saved in demo_output/Middlebury2021/LeRes_2stage/mid21_final, the metric verification is the same as above.(Default base model is set to LeRes50)

python demo.py  --weight checkpoints/model_dict_1_5600.pt \
		--rgb demo_input/Middlebury2021/rgb \
		--twostage