Figure 1 | Scientific Reports

Figure 1

From: Motion illusion-like patterns extracted from photo and art images using predictive deep neural networks

Figure 1

Example images from the dataset and detected optical flows calculated using the Lucas–Kanade method. Four images are shown for each image group: motion illusion, modern arts (painting), realistic arts (painting), movable objects (photograph), and still objects (photograph). Each image group comprises a dataset of 300 images. (a) Original images input to the DNN model [image size: 160 × 120 pixels (width × height)]. (b) Detected optical flow vectors in the predicted images. Yellow points represent the starting points of the detected optical flow vectors, and red lines represent their directions and sizes. The length of the red line is 50 times that of the absolute value of the motion vector calculated using the Lucas–Kanade method. (c) The vectors were drawn over the P1 predicted images.

Back to article page