Attention vs convolution: a survey on OOD object detection
About this project
- Python, Pytorch.
- ImageNet, WordNet.
- Object classification.
- Duration: 1 semester (ongoing).
This project was done in the course CS-503 Visual Intelligence: Machines and Minds at EPFL.
We research the role of the architectures of attention and convolution in the detection of out-of-distribution (OOD) data. We study models that are purely based on convolution (multi-layer perceptrons), purely based on attention (vision-transformers), and other models that mix both.
See our final presentation in the video below:
