This article discusses solving semantic segmentation problems by fine-tuning the KerasCV DeepLabv3+ model.
DeepLabv3+ is a popular semantic segmentation model that can be used in various applications of image segmentation, such as medical imaging, autonomous driving, etc.
KerasCV also integrates DeepLabv3+ into its library. In this blog post, we will discuss extensively how to leverage DeepLabv3+ and fine-tune it based on our custom data. Specifically, we will use the following ImageNet pre-trained backbone as a feature extractor to fine-tune DeepLabv3+:
ResNet50_V2
EfficientNetv2_small
Finally, we will also compare the results of these models.