Time-reversed scene reconstruction from thermal traces using Visual-Language Models and constrained diffusion.
Preprint: a multimodal benchmark and reconstruction framework for inferring recent human-environment interactions from residual physical traces.
Preprint: a multimodal benchmark and reconstruction framework for inferring recent human-environment interactions from residual physical traces.
Time-reversed scene reconstruction from thermal traces using Visual-Language Models and constrained diffusion to recover past frames up to 120 seconds earlier.