Interesting Papers

This is a list of computer vision papers that have caught my eye, but I haven’t yet had a chance to read in detail. Perhaps one might spark inspiration for your project proposal?

VGGT: Visual Geometry Grounded Transformer

ArXiV. This paper appears to solve many problems in geometric computer vision all-in-one.

Fixing the RANSAC Stopping Criterion

ArXiV. This paper claims that somehow we’ve been doing our math wrong on RANSAC all of these years, and there’s an easy way to speed things up.

EVER: Exact Volumetric Ellipsoid Rendering for Real-time View Synthesis

ArXiV. This paper is about a new way to render viewpoints that we haven’t seen if we have a scene that was reconstructed from photos. This is almost a graphics paper.

RePoseD: Efficient Relative Pose Estimation With Known Depth Information

ArXiV. This paper is about single-view depth estimation, like we did in class.

Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass Diffusion Transformers

ArXiV. This paper seems to be a tweak on transformer layers that let vision transformers make higher-resolution output images without an explosion of the parameter count.