Axial-DeepLab - Stand-Alone Axial-Attention for Panoptic Segmentation

Explore a comprehensive video explanation of the research paper "Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation". Delve into the innovative approach of combining learned Positional Embeddings with Axial Attention to create a fully attentional model for image processing. Learn how this new model competes with Convolutional Neural Networks in image classification and achieves state-of-the-art results in various image segmentation tasks. Follow the detailed breakdown of concepts, from the transition from convolution to self-attention in images, to the implementation of axial attention and its application in replacing convolutions in ResNet. Gain insights into the experimental results and practical examples demonstrating the model's effectiveness across multiple large-scale datasets.

- Intro & Overview
- This Paper's Contributions
- From Convolution to Self-Attention for Images
- Learned Positional Embeddings
- Propagating Positional Embeddings through Layers
- Traditional vs Position-Augmented Attention
- Axial Attention
- Replacing Convolutions in ResNet
- Experimental Results & Examples