Overview of our semi-supervised learning framework, S2Contact. (a) The model
is pre-trained on a small annotated dataset. (b) Then, it is deployed on unlabelled
datasets to collect pseudo-labels. The pseudo-labels are filtered with confidence-based
on visual and geometric consistencies. Upon predicting the contact map, the hand and
object poses are optimised to achieve target contact via a contact model.
Video
Abstract
Being able to reason about the physical contacts between
hands and objects is crucial in understanding hand-object manipulation. However, despite the efforts in accurate 3D annotations in hand
and object datasets, there still exist gaps in 3D hand and object reconstructions. Recent works leverage contact maps to refine inaccurate
hand-object pose estimations and generate grasps given object models.
However, they require explicit 3D supervision which is seldom available
and therefore, are limited to constrained settings, e.g., where thermal
cameras observe residual heat left on manipulated objects. In this paper, we propose a novel semi-supervised framework that allows us to
learn contact from monocular videos. Specifically, we leverage visual and
geometric consistency constraints in large-scale datasets for generating
pseudo-labels in semi-supervised learning and propose an efficient graph-
based network to infer contact. Our semi-supervised learning framework
achieves a favourable improvement over the existing supervised learning
methods trained on data with ‘limited’ annotations. Notably, our proposed model is able to achieve superior results with less than half the
network parameters and memory access cost when compared with the
commonly-used PointNet-based approach. We show benefits from using
a contact map that rules hand-object interactions to produce more accurate reconstructions. We further demonstrate that training with pseudo-
labels can extend contact map estimations to out-of-domain objects and
generalise better across multiple datasets.
Framework
A schematic illustration of our framework. We adopt our proposed graph-based network GCN-Contact
as backbone. We utilise a teacher-student mutual learning framework which is composed of a learnable student and an EMA teacher. The student network is trained
with labelled data. For unlabelled data, the student network takes pseudo
contact labels from its EMA teacher and compares with its predictions. (a) refers to contact
consistency constraint for consistency training. To improve the quality of pseudo-label,
we adopt a confidence-based filtering mechanism to geometrically (b) and visually (c)
filter out predictions that violate contact constraints.
Paper and Supplementary Material
Tze Ho Elden Tse, Zhongqun Zhang, Kwang In Kim, Ales Leonardis, Feng Zheng and Hyung Jin Chang S2Contact: Graph-based Network for 3D Hand-Object Contact Estimation with Semi-Supervised Learning
In ECCV, 2022.
This research was supported by the MSIT (Ministry of Science and ICT), Korea, under the ITRC
(Information Technology Research Center) support program (IITP--2022--2020--0--01789) supervised
by the IITP (Institute of Information \& Communications Technology Planning \& Evaluation) and
the Baskerville Tier 2 HPC service (https://www.baskerville.ac.uk/) funded by the Engineering
and Physical Sciences Research Council (EPSRC) and UKRI through the World Class Labs scheme (EP/T022221/1)
and the Digital Research Infrastructure programme (EP/W032244/1) operated by Advanced Research Computing
at the University of Birmingham. KIK was supported by the National Research Foundation of Korea (NRF)
grant (No. 2021R1A2C2012195) and IITP grants (IITP--2021--0--02068 and IITP--2020--0--01336). ZQZ was
supported by China Scholarship Council (CSC) Grant No. 202208060266. AL was supported in part by the EPSRC
(grant number EP/S032487/1). FZ was supported by the National Natural Science Foundation of China under
Grant No. 61972188 and 62122035.
This template was originally made by Phillip Isola and Richard Zhang for a colorful ECCV project; the code can be found here.
