Cut-and-Paste Dataset Generation for Balancing Domain Gaps in Object Instance Detection

Training an object instance detector where only a few training images are available is challenging task. One solution cut-and-paste method that generates dataset by cutting areas out of and pasting them onto other background images. A trained on generated with suffers from the conventional domain shift problem, which stems discrepancy between source (generated dataset) target (real test dataset). Though state-of-the-art adaptation methods able to reduce this gap, it limited because they do not consider difference gaps foreground background. In study, we present gap can be divided into two sub-domain for Then, show original approach new unbalanced gaps, has separate domains background, unlike problem. introduce advanced balance diversifying GAN (generative adversarial network)-generated seed simplifying using image processing techniques. Experimental results our effective balancing improving accuracy detection in cluttered indoor environment Furthermore, improve methods.