Addressing Source Scale Bias via Image Warping for Domain Adaptation

arXiv:2403.12712v1 Announce Type: cross
Abstract: In visual recognition, scale bias is a key challenge due to the imbalance of object and image size distribution inherent in real scene datasets. Conventional solutions involve injecting scale invariance priors, oversampling the dataset at different scales during training, or adjusting scale at inference. While these strategies mitigate scale bias to some extent, their ability to adapt across diverse datasets is limited. Besides, they increase computational load during training and latency during inference. In this …

abstract adjusting arxiv bias challenge cs.cv cs.lg dataset datasets distribution domain domain adaptation image inference key object oversampling recognition scale solutions training type via visual