石 (Rock), 财 (Wealth), 壶 (Kettle), 笔 (Writing brush), 松 (Pine) (2019) is a video installation that references the tradition of East Asian ‘still life’ paintings. To create the video, thousands of photographs of the five objects and plants were taken. Those images were processed and trained on with an image classification neural network (VGGNet60). The resulting small images were combined through enlarging and blending to form a higher resolution image. The larger images are then recolored through Pix2pix, a Generative Adversarial Network (GAN). Finally, the recolored images were interpolated between frames to form the video.
Machine learning neural networks that involve images as datasets accepts images taken from multiple perspectives and scale. The prime characteristic of “scattered perspective” is in its dynamic perspectival quality. “Scattered perspective” was utilized as a way of visual depiction in paintings from the Wei and Jin dynasties in China. Paintings that utilize scattered perspective do not depict a scene seen from a fixed point towards a fixed direction. The neural network learns the distinguishing features of the category without taking into account the perspectival or scalar properties of the features. 石 (Rock), 财 (Wealth), 壶 (Kettle), 笔 (Writing brush), 松 (Pine) aims to draw a parallel between the processing method of neural networks and scattered perspective.
More on this project: