bioRxivJClub
Natural scene segmentation dynamics reveal iterative Bayesian inference
bioRxiv · · Preprint · Open access
Biswas, Vacher + 3 more
Abstract ↗AI summary
The abstract is read at the publisher; the summary is JClub's.
Iterative Bayesian inference for natural scene segmentation predicts and matches participants’ segmentation tempo, showing faster inference when visual inputs align with a local smoothness prior.
- Why it matters: Natural scene segmentation is ambiguous and subjective, leaving the spatiotemporal dynamics of how perceptual objects are formed poorly explained by existing theories.
- What they did: Humans produced subjective segmentation maps for natural images using an innovative paradigm, while Iterative Bayesian Inference algorithms iteratively combined visual inputs with a local smoothness prior.
- The result: When inputs were consistent with the spatial prior, iterative inference was faster, and this prior–tempo relationship was evident in the data and reflected each participant’s spatial biases.
The findingWhy it mattersWhat they didThe result
- Open access