Sub-pixel Detection

1. Project description

Sub-pixel detection project in SJTU CS386. Our object is to detect interest points in sub-pixel accuracy. We proposed SPResNet, using residual blocks, upsampling and pixel shuffle to extract sub-pixel points.

The implementation is based on Pytorch. Experiments runs on MacOS.

2. Usage

2.1. install requirements

pip install -r requirements.txt

2.2. prepare datasets

Put train HR(High-Resolution) images in ./data/intput/train_input, put validation HR images in ./data/input/valid_input, and (optionally) put testing HR images in ./data/input/test_input. We already put some demo images in these folders.

2.3. train and test

The interface is main.py. The following command will run the program by default settings. Defaultly it will pre-process the data, then train and test.

python main.py

Check ./utils/option.py for arguments details. Or running help:

python main.py -h

2.4. Output

Output of test is in ./data/output/test_output.

3.remarks

3.1. Dataset download

We downloaded DIV2K dataset from https://github.com/xinntao/BasicSR/wiki/Prepare-datasets-in-LMDB-format . Other datasets are also accpetable.

3.2. Image type

By default, only .png images are valid input. If other type of images need to be supported, please modify the following code in ./utils/dataPrepare.py at Line14.

if name.endswith(".png"):

FengMingquan-sjtu / spd