DeePaste -- Inpainting for Pasting

TL;DR

DeePaste introduces an inpainting-based method for object pasting that reduces artifacts and improves the quality of synthetic datasets, leading to better performance on real-world tasks like detection and segmentation.

Contribution

The paper presents a novel inpainting technique for seamless object pasting, enhancing synthetic data quality for training models.

Findings

Achieves state-of-the-art results on instance detection

Improves foreground segmentation performance

Reduces artifacts in pasted objects

Abstract

One of the challenges of supervised learning training is the need to procure an substantial amount of tagged data. A well-known method of solving this problem is to use synthetic data in a copy-paste fashion, so that we cut objects and paste them onto relevant backgrounds. Pasting the objects naively results in artifacts that cause models to give poor results on real data. We present a new method for cleanly pasting objects on different backgrounds so that the dataset created gives competitive performance on real data. The main emphasis is on the treatment of the border of the pasted object using inpainting. We show state-of-the-art results both on instance detection and foreground segmentation