Concept

"TRAN-SCEPTION : de/re-constructing our segmented reality" is a group project realised during DfPI 3-day design workshop “Mapping the Machine” at Bartlett, UCL. It is an early-stage exploration of machine learning techniques used for creative practice. We create maps of machinic space to discover their boundaries and potentially begin navigating these worlds as reflections of our own biases and limitations.

In this demo, we use footages from the movie Inception (one that metaphorically matches the theme of our work), deconstruct the images into colour blocks of different categories of objects and then reconstruct another segmented “reality”. The process is like to dream unconsciously with machinic views. The output is oddly familiar.

Technical details

The #machinelearning project is based on Runway ML pre-trained models of Spade-COCO, DeepLab and DeepPrivacy. The datasets used for training are mainly COCO-Stuff 10k/164k and PASCAL VOC 2012.

概念

"TRAN-SCEPTION:解構/重構我們的碎片現實"是一個在巴特萊特學院互動建築實驗室3天設計研討會「機器映射」中實現的小組專案。它是對用於創造性實踐的機器學習技術的早期探索。我們創建機械空間映射來發現它們的界限,並藉此重新審視這個世界如何反射出我們的個人偏見和局限性。

在此演示中,我們使用電影《盜夢空間》中的素材(隱喻地與我們的工作主題相匹配),將圖像解構成不同類別物件的色塊,然後重建另一個碎片化的“現實"。這個過程就像無意識地用機械的觀點做夢。成果古怪而似曾相識。

技術細節

這個機器學習項目基於Runway ML預訓練模型Spade-COCO, DeepLab和DeepPrivacy。訓練所用的數據庫主要為COCO-Stuff 10k/164k和PASCAL VOC 2012。

Trans-ception

Related posts :