VAPNet-Pytorch

This is an unofficial Pytorch implementation of View Adjustment Prediction Model in Camera View Adjustment Prediction for Improving Image Composition(2021).

View Adjustment Prediction Model

View Adjustment Prediction Model, which we named as VAPNet, predicts whether an image needs adjustment, which adjustment, and how much adjustment is needed.

Performance

	AUC	TPR	Left	Right	Up	Down	Zoom-in	Zoom-out	Clockwise	Counter-Clockwise	IoU
Original Paper	0.608	0.436	0.221	0.221	0.390	0.341	0.015	0.378	0.124	0.110	0.750
Version1	0.624	0.483	0.263	0.240	0.332	0.295	0.015	0.061	0.152	0.144	0.750
Version2	0.627	0.485	0.324	0.300	0.494	0.512	None	None	None	None	0.759

Columns from Left to Counter-Clockwise means F1-Score

Dataset

GAICD [Paper][Github]
FCDB [Paper][Github]
Open Images[Homepage][CVDF]

Usage

git clone https://github.com/PROLCY/VAPNet-Pytorch.git
cd VAPNet-Pytorch && mkdir weight

pip install -r requirements.txt

Demo

Download pretrained model in the directory weight

	Adjustment	IoU	Download
Version1	Left, Right, Up, Down, Zoom-in, Zoom-out, Clockwise, Counter-Clockwise	0.750	Link
Version2	Left, Right, Up, Down	0.759	Link

python demo.py {image_dir_path}

You can check the inference result in terminal.

Result

Predicted view adjustment is as follows.

If you are interested in this repository, please contact ckd248@naver.com