Freelance - Pytorch, GAT

vision
1

  1. There is such an article ([2110.01200] AASIST: Audio Anti-Spoofing using Integrated Spectro-Temporal Graph Attention Networks), it uses the graph attention network (GAT) to detect voice forgery. Here you need to pay attention to Fig 1. In fact, you need to make such an algorithm, but not for voice, but for faces.

  2. This circuit itself (Fig 1) has an Encoder. You can take ResNet18 as it, but without the last layer - average polling + FC.

  3. Next, you need to take the GAT itself. For example, here is such an implementation (GitHub - Diego999/pyGAT: Pytorch implementation of the Graph Attention Network model by Veličković et. al (2017, https://arxiv.org/abs/1710.10903)).

  4. Next, take some database of faces. It is important that the size of the images fit ResNet18, otherwise you will need to resize.

  5. Put it all together. That is, a face is given at the input, then ResNet18 comes, then GAT. Then you need to look at the diagram, but if I’m not mistaken, then there is a certain vector, then FC and human marks (I can describe in more detail in the messages).

  6. Train

You can offer your price.

Or help find a person who will help with the implementation

2

I would recommend to change the category to “Jobs” so that more interested people would see it.