Convolution Neural Networks have made great strides in accurately labeling images. But they require huge amounts of training data. But there are only so many number of these images on the internet that can be found. To further restrict our dataset, these images are often subject to copyrights. Even if we are able to get a big enough training corpus, we still need to augment the images. By augmentation, I mean if an image of a desert is flipped upside down, a good model should still be able to detect it.
In this project, an image processing pipeline is made. The pipeline automatically searches for all the images in a folder and applies various image augmentations like resizing, graining, flipping, and so on.
Sample Input
Sample Output
And so on...
| Visit: | Project link |