Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Neural Implicit Representations for Physical Parameter Inference from a Single Video

About

Neural networks have recently been used to analyze diverse physical systems and to identify the underlying dynamics. While existing methods achieve impressive results, they are limited by their strong demand for training data and their weak generalization abilities to out-of-distribution data. To overcome these limitations, in this work we propose to combine neural implicit representations for appearance modeling with neural ordinary differential equations (ODEs) for modelling physical phenomena to obtain a dynamic scene representation that can be identified directly from visual observations. Our proposed model combines several unique advantages: (i) Contrary to existing approaches that require large training datasets, we are able to identify physical parameters from only a single video. (ii) The use of neural implicit representations enables the processing of high-resolution videos and the synthesis of photo-realistic images. (iii) The embedded neural ODE has a known parametric form that allows for the identification of interpretable physical parameters, and (iv) long-term prediction in state space. (v) Furthermore, the photo-realistic rendering of novel scenes with modified physical parameters becomes possible.

Florian Hofherr, Lukas Koestler, Florian Bernard, Daniel Cremers• 2022

Related benchmarks

TaskDatasetResultRank
Parameter EstimationPendulum 90cm
Length (m)0.84
9
Parameter EstimationDelfys Pendulum 45cm
Length (m)0.77
6
Parameter EstimationDelfys Torricelli Small
k [sqrt(m)/s]0.21
6
Parameter EstimationDelfys Torricelli Med
Parameter k (sqrt(m)/s)0.14
6
Parameter EstimationDelfys Torricelli Large
k [sqrt(m)/s]0.16
6
Parameter EstimationDelfys Pendulum 150cm
Length (m)0.63
6
Parameter EstimationDelfys Sliding Block Low
Acceleration (m/s^2)-0.09
6
Parameter EstimationDelfys Sliding Block Med
Acceleration (m/s^2)-0.01
6
Parameter EstimationDelfys Sliding Block High
Estimated Acceleration (m/s^2)-0.06
6
String-length recoveryReal-video pendulum L* = 0.45 m (test)
Estimated Length0.77
4
Showing 10 of 12 rows

Other info

Follow for update