DeepPano

Necessary Setup

make following directories

• DeepPano/data/
• DeepPano/data/metadata/
• DeepPano/data/rawdata/
• DeepPano/data/rawdata/panoImg: original panorama images
• DeepPano/data/rawdata/psdFile: annotation files
• DeepPano/data/rawdata/xmlFile: PanoSeg or box information
• DeepPano/result/
• DeepPano/result/checkpoint: for postprocess

To Generate FirstFile.csv

python3 firstfilegen.py semi

it iterates DeepPano/data/rawdata/panoImg files to generate FirstFile for all panorama images

and generate DeepPano/data/metadata/SemiSet.csv

python3 firstfilegen.py not

it iterates DeepPano/data/rawdata/psdFile files to generate FirstFile for all annotated panorama images

and generate DeepPano/data/metadata/DataSet.csv

or you can use your own .csv file that has Image.Title, Pano.File, Xml.File, Annot.File, Train.Val columns

To Generate Image Patches And According .csv

python3 manage.py genData 10 (FirstFile.csv route)

This generates image patches in DeepPano/data/metadata/(firstfile name)/

and metadata of DeepPano/data/metadata/(firstfile name)-10.csv

To use this dataset for training, you need to make DeepPano/data/StatDataset.csv

python3 manage.py genStat (SecondFile.csv route)

It generates pano/box mean/std stat for the dataset

To Generate Result

Necessary Setup

put checkpoint under DeepPano/result/checkpoint/(directory name)/checkpoint/

and config file(.json) used to train that checkpoint at DeepPano/result/checkpoint/(directory name)/

SecondFile.csv and its Dataset, and its panoImgs should also be prepared

python3 postprocess.py -1 (SecondFile.csv route)

This iterates all folders in DeepPano/result/checkpoint/ and generate result for all image patches in SecondFile.csv

python3 postprocess.py (checkpoint directory route) (SecondFile.csv route)

This iterates all checkpoints in checkpoint directory and generate result for all image patches in SecondFile.csv