- Reproduced and confirmed the results from above NIPS 2019 research paper for course project.
- Implemented ResNet50 convolution network with self-attention layers as per the guidelines in the paper.
- Analyzed effect of spatial extent, positional embedding, and attention type on ResNet50 attention model.
- Implemented SSD object detection model by replacing set of convolution layers with self-attention layers.
- Evaluated above SSD model for performance change in terms of Accuracy (4.4% less), number of model parameters (5.6% less), and Flops count (very less).
lilujunai / stand-alone-self-attention-in-vision-models- Goto Github PK
View Code? Open in Web Editor NEWThis project forked from akashpatil-pixel32/stand-alone-self-attention-in-vision-models-
Reproduced and confirmed the results from above NIPS 2019 research paper for course project.