Human action recognition using machine learning
The accurate detection of human pose is useful in various areas such as surveillance, clothes parsing and game modelling. While many existing models attempt to breach the gap, they suffer from problems including learning uninformative latent variables and being memory- intensive. In this work, we...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/175048 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | The accurate detection of human pose is useful in various areas such as surveillance, clothes
parsing and game modelling. While many existing models attempt to breach the gap, they
suffer from problems including learning uninformative latent variables and being memory-
intensive. In this work, we propose FFPoser, a diffusion-based model that leverages the use
of Fast-Fourier Transformation and a complex-valued neural network to resolve the above
issues. It is able to generate fairly realistic poses and produce accurate occlusion recovery
results. Additionally, it is easy to train while still be able to learn the important information
of the source data. |
|---|