TNET: A NEURAL NETWORK FOR BLURRING BACKGROUND

DATASET

TRAIN

We use AiFenGe dataset for training. You can download it from the link: https://www.kaggle.com/laurentmih/aisegmentcom-matting-human-datasets/

TEST

We use EBB! dataset for testing. There are train, val and test data in the dataset. However, images with blurred background are in the train dataset only, so we choose EBB!-train dataset for testing TNET. Please download it from the link: https://competitions.codalab.org/competitions/24716#participate

REQUIREMENTS

Pytorch 1.17

Python 3.6

PROCESS

1. Generate The Dataset

python3 preprocess.py

2. Train

python3 train.py

3. Blur The Image

python3 blur_image.py

4. Generate The Image With A Blurred Background

python3 blurred_background.py

TEST RESULTS

We use peak signal-noise ratio (PSNR) and structual similarity (SSIM) as the evaluation indicators.

PSNR	SSIM
21.92	0.74

REFERENCE

https://qianbin.blog.csdn.net/article/details/105787453