Skip to content

Latest commit

 

History

History
40 lines (31 loc) · 1021 Bytes

README.md

File metadata and controls

40 lines (31 loc) · 1021 Bytes

Dataset Preparation

1. Generate foreground object and mask

We use PASCAL VOC dataset for the foreground.

The data is organised as follows:

├── background
└── VOC2010
    ├── Annotations
    ├── ImageSets
    │   ├── Action
    │   ├── Layout
    │   ├── Main
    │   └── Segmentation
    ├── JPEGImages
    ├── SegmentationClass
    └── SegmentationObject

Run this to generate foreground.

python generate_foreground.py

2. Prepare Background

We crop DIV2K dataset to 256*256 for background images.

3. Generate Datasets

Run this to generate RGB datasets (./trainset0908).

python generate_motion.py --config ./config.yml

  • step: the original images with sub-pixel motions

  • gt: the ground-truth of the alpha times maginified images

  • The number behind underscore is magnification factor.