Franca
Learning visual features at several levels of detail for dense prediction.
Franca learns visual features at several levels of detail, improving the features used for dense prediction tasks.
I’m a co-first author. I worked on the research ideas and code, trained the model at standard and high resolutions, ran experiments and ablations, and helped write the paper.
I designed and implemented RASA, Franca’s method for removing unwanted absolute-position information from visual features. I also ran its experiments and ablations.