notesum.ai
Published at October 30TPP-Gaze: Modelling Gaze Dynamics in Space and Time with Neural Temporal Point Processes
cs.CV
cs.AI
Released Date: October 30, 2024
Authors: Alessandro D'Amelio1, Giuseppe Cartella2, Vittorio Cuculo2, Manuele Lucchi1, Marcella Cornia2, Rita Cucchiara2, Giuseppe Boccignone1
Aff.: 1University of Milan, Italy; 2University of Modena and Reggio Emilia, Italy

| Dim | GMM | MM (KL-Div) | SM (KL-Div) | SED | |||||||||||
| CNN | Img | TPP | Dur | Avg | w/ Dur | w/o Dur | Avg | ||||||||
| Image Backbone | |||||||||||||||
| RN | 256 | 256 | 4 | 16 | 0.011 | 0.037 | 0.113 | 0.101 | 17.575 | ||||||
| DN | 256 | 256 | 4 | 16 | 0.012 | 0.028 | 0.078 | 0.060 | 17.032 | ||||||
| Image and TPP Dimensionalities | |||||||||||||||
| DN | 128 | 128 | 4 | 16 | 0.010 | 0.031 | 0.094 | 0.069 | 16.959 | ||||||
| DN | 128 | 256 | 4 | 16 | 0.012 | 0.030 | 0.084 | 0.063 | 16.887 | ||||||
| DN | 256 | 128 | 4 | 16 | 0.009 | 0.037 | 0.105 | 0.082 | 17.413 | ||||||
| DN | 256 | 256 | 4 | 16 | 0.012 | 0.028 | 0.078 | 0.060 | 17.032 | ||||||
| DN | 256 | 512 | 4 | 16 | 0.010 | 0.031 | 0.101 | 0.095 | 17.462 | ||||||
| DN | 512 | 256 | 4 | 16 | 0.008 | 0.032 | 0.110 | 0.093 | 17.497 | ||||||
| DN | 512 | 512 | 4 | 16 | 0.009 | 0.027 | 0.104 | 0.077 | 17.154 | ||||||
| Mixture Components | |||||||||||||||
| DN | 256 | 256 | 2 | 16 | 0.014 | 0.027 | 0.092 | 0.071 | 16.944 | ||||||
| DN | 256 | 256 | 4 | 16 | 0.012 | 0.028 | 0.078 | 0.060 | 17.032 | ||||||
| DN | 256 | 256 | 2 | 32 | 0.009 | 0.030 | 0.109 | 0.098 | 17.216 | ||||||
| DN | 256 | 256 | 4 | 32 | 0.009 | 0.031 | 0.103 | 0.076 | 17.252 | ||||||