Breaking01 / AI
FreeDeshadow targets shadows cast by selected objects
The framework lets users select an object with a dot, line, mask or text to remove only the shadow it casts.

Researchers submitted a paper on FreeDeshadow, a framework for removing shadows cast by user-selected objects, alongside a dataset containing 154,571 sets of four items, according to the paper.
The paper, submitted on 10 Nov, describes how users select a target object with a dot, line, object mask or text. Text prompts describe the object's attributes, spatial relationships or state.
According to the abstract, the PCSRD+ dataset contains images at 1024×1024 resolution and supports training and evaluation.
The researchers said in the paper that they designed a generator to interpret the user's input and a separate refiner to limit changes to the target shadow. This design aims to remove that shadow while preserving other content.
The paper describes earlier methods as either removing all shadows or relying on precise shadow masks, which mark the areas to be edited. FreeDeshadow instead lets users specify the object casting the shadow.
The abstract does not disclose numerical performance differences between FreeDeshadow and earlier methods.