Summary of Scaling Graph Convolutions For Mobile Vision, by William Avery et al.
Scaling Graph Convolutions for Mobile Vision
by William Avery, Mustafa Munir, Radu Marculescu
First submitted to arxiv on: 9 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Mobile Graph Convolution (MGC) is introduced as a new vision graph neural network (ViG) module to solve the scaling problem of MobileViG, which falls behind models with similar latency. MGC improves on Sparse Vision Graph Attention (SVGA) by increasing graph sparsity and introducing conditional positional encodings to the graph operation. The proposed mobile vision architecture, MobileViGv2, uses MGC to achieve state-of-the-art results in image classification, object detection, instance segmentation, and semantic segmentation tasks. The smallest model, MobileViGv2-Ti, achieves a top-1 accuracy of 77.7% on ImageNet-1K with 0.9 ms inference latency, while the largest model, MobileViGv2-B, achieves an 83.4% top-1 accuracy with 2.7 ms inference latency. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Mobile Graph Convolution (MGC) is a new module that helps solve a problem in mobile vision architecture called MobileViG. It makes it possible to have a bigger and better model without slowing down the phone too much. The new architecture, called MobileViGv2, uses MGC to do well on many tasks like classifying images, finding objects, and segmenting scenes. The smallest model does well with just 0.9 milliseconds of computing time, while the biggest model does even better with 2.7 milliseconds. |
Keywords
» Artificial intelligence » Attention » Graph neural network » Image classification » Inference » Instance segmentation » Object detection » Semantic segmentation
