FocalDB: A Dataset for Autofocusing Projectors Using Images and Videos with Varying Levels of Focus

Citation Author(s):
Soham
Patil
K. J. Somaiya College Of Engineering, Vidyavihar, Somaiya Vidyavihar University, Mumbai, India
Ninad
Mehendale
K. J. Somaiya College Of Engineering, Vidyavihar, Somaiya Vidyavihar University, Mumbai, India
Submitted by:
Ninad Mehendale
Last updated:
Tue, 03/21/2023 - 01:06
DOI:
10.21227/5knt-hr48
Data Format:
Research Article Link:
License:
0
0 ratings - Please login to submit your rating.

Abstract 

The Autofocus Projector Dataset is a collection of 555 images and 150 videos captured while projecting images and videos with varying levels of Gaussian blur. The dataset includes images and videos of different blur levels, ranging from fully focused to the maximum levels of left and right Gaussian blur as per the projector's specifications. The dataset was recorded using a Redmi Note 11T 5G mobile camera with a 50 MP, f/1.8, 26mm (wide) sensor, PDAF image camera, and 1080p@30 fps video camera. The images and videos are organized into 111 and 30 folders, respectively, with each folder containing 5 images or videos. Each folder includes 4 images or videos with different blur levels and a clear, sharp image or video of the same image or video that was blurred at various levels. The index files for the images and videos contain metadata such as name, date, and time of capture, blur level, format, size, and dimensions. This dataset can be used to develop and train algorithms for autofocus in projectors.

Instructions: 

 Extract the zip(s) of the dataset either of images or videos as per your requirement.

 

·        After extracting a folder containing all the folders in the dataset should be contained in the extracted directory. The image dataset contains 111 folders and 1 index file and the video dataset contains 30 folders and 1 index file.

 

·        Don’t rename or rearrange the files or folders; doing such will result in disparity between the index file and the dataset.

 

·        You can then use the dataset according to the purpose of yours.

 

·        Like, if you want to train a neural network to detect blurs or correct blur, divide the dataset into training and testing data.

 

·        Then classify the input and result or output to that input; you can do this using index file which shows the classification of the blurred images and a clear sharp image for the same blurred images.

 

·        Then using a Python script you can train the dataset and test the network on the remaining dataset.