Finding sky pixels in an image using pre-trained PSPNet segmentation model
The model takes input as an image containing some sky area and marks the sky area in white and other areas in black. The training data consists of hundreds of images taken from a fixed perspective at a location and a mask image denoting the sky pixels in white and other areas of the image in black. The model gives an output image which marks the sky area in white and other areas in black. The data is downloaded and stored as:-
├── 1093
│ ├── 1093_mask.png
│ ├── 20140418_215418.jpg
│ ├── 20140419_002403.jpg
│ └── .....
├── 10066
│ ├── 10066_mask.png
│ ├── 20130101_084634.jpg
│ ├── 20130101_091628.jpg
│ └── .....
├── ....
├── SkyPixel.ipynb
├── SkyPixelPredict.ipynb
└── skyPixelModel.h5
The images were trained after scaling down the resolution to 240 * 240 pixels. The predicted mask image will also be in the same resolution.
A sample test on the trained model:-
