English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

General physical scene understanding requires more than simply localizing andrecognizing objects -- it requires knowledge that objects can have differentlatent properties (e.g., mass or elasticity), and that those properties affectthe outcome of physical events. While there has been great progress in physicaland video prediction models in recent years, benchmarks to test theirperformance typically do not require an understanding that objects haveindividual physical properties, or at best test only those properties that aredirectly observable (e.g., size or color). This work proposes a novel datasetand benchmark, termed Physion++, that rigorously evaluates visual physicalprediction in artificial systems under circumstances where those predictionsrely on accurate estimates of the latent physical properties of objects in thescene. Specifically, we test scenarios where accurate prediction relies onestimates of properties such as mass, friction, elasticity, and deformability,and where the values of those properties can only be inferred by observing howobjects move and interact with other objects or fluids. We evaluate theperformance of a number of state-of-the-art prediction models that span avariety of levels of learning vs. built-in knowledge, and compare thatperformance to a set of human predictions. We find that models that have beentrained using standard regimes and datasets do not spontaneously learn to makeinferences about latent properties, but also that models that encode objectnessand physical states tend to make better predictions. However, there is still ahuge gap between all models and human performance, and all models' predictionscorrelate poorly with those made by humans, suggesting that no state-of-the-artmodel is learning to make physical predictions in a human-like way. Projectpage: https://dingmyu.github.io/physion_v2/

Physion++: Evaluating Physical Scene Understanding that Requires Online Inference of Different Physical Properties