In recent years, neural methods have become widespread in camera-based reconstructions. In most cases, however, hundreds of camera perspectives are needed. Meanwhile, conventional photometric methods exist which can compute highly precise reconstructions even from objects with textureless surfaces. However, these typically work only under controlled lab conditions.

Daniel Cremers, professor of Computer Vision and Artificial Intelligence at TUM and leader of the Munich Center for Machine Learning (MCML) and a director of the Munich Data Science Institute (MDSI) has developed a method together with his team that utilizes the two approaches.

It combines a neural network of the surface with a precise model of the illumination process that considers the light absorption and the distance between the object and the light source. The brightness in the images is used to determine the angle and distance of the surface relative to the light source.

“That enables us to model the objects with much greater precision than existing processes. We can use the natural surroundings and can reconstruct relatively textureless objects for our reconstructions,” says Cremers.

The paper is published on the arXiv preprint server and will be presented at the Conference on Computer Vision and Pattern Recognition (CVPR 2024) held in Seattle from June 17 to June 21, 2024.

Applications in autonomous driving and preservation of historical artifacts

The method can be used to preserve historical monuments or digitize museum exhibits. If these are destroyed or decay over time, photographic images can be used to reconstruct the originals and create authentic replicas.

The team of Prof. Cremers also develops neural camera-based reconstruction methods for autonomous driving, where a camera films the vehicle’s surroundings. The autonomous car can model its surroundings in real-time, develop a three-dimensional representation of the scene, and use it to make decisions.

The process is based on neural networks that predict 3D point clouds for individual video images that are then merged into a large-scale model of the roads traveled.

A small team of AI researchers at Google’s DeepMind project has found that LLMs are not very good at writing jokes that are funny. They asked stand-up comedians to use LLMs to write a stand-up routine for them and posted their findings on the arXiv preprint server.

To be successful, most stand-up comedians have to write a stand-up routine and perform it on stage. Such routines, or monologs, typically involve both storytelling and jokes, or describe humorous situations. Many also employ surprising or incongruous remarks, giving the audience a sudden insight into something they may not have considered in a certain way before.

Most professional stand-up comedians spend a great deal of time polishing their routines and testing them on small audiences before performing in front of large crowds or on television specials.

In this new effort, the team at DeepMind wondered if LLMs might be capable of creating not just jokes, but entire stand-up routines. To find out, they recruited 20 professional stand-up comedians who had used LLMs in their work before. The performers used an LLM to help them write an entire routine and then rated the results.

The researchers found that LLMs were quite good at coming up with jokes; unfortunately, few if any were funny. Most, they suggested, were generic in nature and few offered anything in the way of a surprise.

Overall, the comedians found the AI-generated jokes to lack the cutting edge that is typically needed for a joke to be funny. Many described the results as bland. But some of them did find the LLM to be useful in generating a routine that could be used for creating a basic structure around which they could build their own jokes.

The research team suggests the results were not surprising considering that makers of LLMs use filters to prevent them from generating output that could be offensive or edgy.

© 2024 Science X Network