Anton Milan
Anton Milan (né Andriyenko)
I am doing research in machine learning, computer vision and robotics at the Amazon Research and Development Center in Berlin. You can find more information about me here.
  • 11/01/2019 - PAMI paper on RefineNet accepted.
  • 21/12/2018 - We will hold another MOTChallenge Workshop at CVPR 2019 in Los Angeles.
  • 19/02/2018 - PoseTrack paper accepted to CVPR.
  • 15/01/2018 - Two ICRA papers accepted on ARC system and perception.
  • 09/11/2017 - AAAI'18 paper accepted.
  • 15/08/2017 - I joined the Amazon Core Machine Learning Group in Berlin.
  • 30/07/2017 - Our team won the Final Round at the Amazon Robotics Challenge.

See also Google Scholar

2019

RefineNet: Multi-Path Refinement Networks
for Dense Prediction

G. Lin, F. Liu, A. Milan, C. Shen, I. Reid. In PAMI 2019
bibtex | abstract | paper | code

@article{Lin:2019:PAMI,
        Author = {Lin, G. and Liu, F. and Milan, A. and Shen, C. and Reid, I.},
	Title = {Refine{N}et: {M}ulti-Path Refinement Networks for Dense Prediction},
	volume = {41},
	number = {1},
	month={Jan},
	pages={1-1},
	doi={10.1109/TPAMI.2019.2893630},
	journal = {IEEE TPAMI},
	year = {2019}
}
			
Recently, very deep convolutional neural networks (CNNs) have shown outstanding performance in object recognition and have also been the first choice for dense prediction problems such as semantic segmentation and depth estimation. However, repeated sub-sampling operations like pooling or convolution striding in deep CNNs lead to a significant decrease in the initial image resolution. Here, we present RefineNet, a generic multi-path refinement network that explicitly exploits all the information available along the down-sampling process to enable high-resolution prediction using long-range residual connections. In this way, the deeper layers that capture high-level semantic features can be directly refined using fine-grained features from earlier convolutions. The individual components of RefineNet employ residual connections following the identity mapping mindset, which allows for effective end-to-end training. Further, we introduce chained residual pooling, which captures rich background context in an efficient manner. We carry out comprehensive experiments on semantic segmentation which is a dense classification problem and set new state-of-the-art results on seven public datasets. We further apply our method for depth estimation and demonstrate the effectiveness of our method on dense regression problems.
			

2018

PoseTrack: A Benchmark for Human Pose Estimation and Tracking

Iqbal, U. and Milan, A. and Andriluka, M. and Ensafutdinov, E. and Pishchulin, L. and Gall, J. and Schiele B. In CVPR 2018
bibtex | abstract | paper |

@inproceedings{PoseTrack,
	Title = {Pose{T}rack: {A} Benchmark for Human Pose Estimation and Tracking},
	booktitle = {CVPR},
	Author = {Iqbal, U. and Milan, A. and Andriluka, M. and Ensafutdinov, E. and Pishchulin, L. and Gall, J. and Schiele B.},
	Year = {2018}
}
			
Human poses and motions are important cues for analysis of videos
with people and there is strong evidence that representations based
on body pose are highly effective for a variety of tasks such as
activity recognition, content retrieval and social signal processing.
In this work, we aim to further advance the state of the art by
establishing "PoseTrack" , a new large-scale benchmark for video-based
human pose estimation and articulated tracking, and bringing together
the community of researchers working on visual human analysis. The
benchmark encompasses three competition tracks focusing on
i) single-frame multi-person pose estimation, ii) multi-person pose
estimation in videos, and iii) multi-person articulated tracking.
To facilitate the benchmark and challenge we collect, annotate and
release a new %large-scale benchmark dataset that features videos
with multiple people labeled with person tracks and articulated pose.
A centralized evaluation server is provided to allow participants to
evaluate on a held-out test set. We envision that the proposed benchmark
will stimulate productive research both by providing a large and
representative training dataset as well as providing a platform to
objectively evaluate and compare the proposed methods. The benchmark
is freely accessible at posetrack.net.
			

Semantic Segmentation from Limited Training Data

A. Milan et al. In ICRA 2018
bibtex | abstract | paper | video

@inproceedings{Milan:2018:ICRA,
	Title = {Semantic Segmentation from Limited Training Data},
  Booktitle = {ICRA}
	Author = {Milan, A. and Pham, T. and Vijay, K. and Morrison, D and Tow, A. W. and Liu, L. and
    Erskine, J. and Grinover, R. and Gurman, A. and Hunn, T. and Kelly-Boxall, N. and Lee, D. and
    McTaggart, M. and Rallos, G. and Razjigaev, A. and Rowntree, T. and Shen T. and Smith, R. and
    Wade-McCue, S. and Zhuang, Z. and Lehnert, C. and Lin, G. and Reid, I. and Corke, P. and Leitner, J.},
  Month = {May},
  Year = {2018}
}
			
We present our approach for robotic perception in cluttered scenes that
led to winning the recent Amazon Robotics Challenge (ARC) 2017. Next to
small objects with shiny and transparent surfaces, the biggest challenge
of the 2017 competition was the introduction of unseen categories. In
contrast to traditional approaches which require large collections of
annotated data and many hours of training, the task here was to obtain
a robust perception pipeline with only few minutes of data acquisition
and training time. To that end, we present two strategies that we explored.
One is a deep metric learning approach that works in three separate steps:
semantic-agnostic boundary detection, patch classification and pixel-wise
voting. The other is a fully-supervised semantic segmentation approach with
efficient dataset collection. We conduct an extensive analysis of the two
methods on our ARC 2017 dataset. Interestingly, only few examples of each
class are sufficient to fine-tune even very deep convolutional neural
networks for this specific task.
			

Cartman: The low-cost Cartesian Manipulator that won the Amazon Robotics Challenge

D. Morrison et al. In ICRA 2018
bibtex | abstract | paper | video

@inproceedings{Morrison:2018:ICRA,
	Title = {Cartman: {T}he low-cost Cartesian Manipulator that won the {A}mazon {R}obotics {C}hallenge},
  booktitle = {ICRA}
	Author = {Morrison, D. and others},
  month = {May},
  year = {2018}
}
			
The Amazon Robotics Challenge enlisted sixteen teams to each design a
pick-and-place robot for autonomous warehousing, addressing development
in robotic vision and manipulation. This paper presents the design of our
custom-built. cost-effective robot system Cartman, which won first place
in the competition finals by stowing 14 (out of 16) and picking all 9 items
 in 27 minutes, scoring a total of 272 points. We highlight our experience-
 centred design methodology and key aspects of our system that contributed
 to our competitiveness. We believe these aspects are crucial to building
 robust and effective robotic systems.
			

Joint Learning of Set Cardinality and State Distribution

S. H. Rezatofighi, A. Milan, Q. Shi, A. Dick, I. Reid. In AAAI
bibtex | abstract | paper |

@inproceedings{Rezatofighi:2018:JDS,
	Title = {Joint Learning of Set Cardinality and State Distribution},
	url = {https://arxiv.org/abs/1709.04093},
	booktitle = {AAAI},
	Author = {Rezatofighi, S. H. and Milan, A. and Shi, Q. and Dick, A. and Reid, I.},
	Year = {2018}
}
			
We present a novel approach for learning to predict sets using deep learning.
In recent years, deep neural networks have shown remarkable results in computer
vision, natural language processing and other related problems. Despite their success,
traditional architectures suffer from a serious limitation in that they are built to
deal with structured input and output data, i.e. vectors or matrices. Many real-world
problems, however, are naturally described as sets, rather than vectors. Existing
techniques that allow for sequential data, such as recurrent neural networks,
typically heavily depend on the input and output order and do not guarantee a valid
solution. Here, we derive in a principled way, a mathematical formulation for set
prediction which is permutation invariant. In particular, our approach jointly learns
both the cardinality and the state distribution of the target set. We demonstrate the
validity of our method on the task of multi-label image classification and achieve a new
state of the art on the PASCAL VOC and MS COCO datasets.
			


2017

DeepSetNet: Predicting Sets with Deep Neural Networks

S. H. Rezatofighi, V. Kumar, A. Milan, E. Abbasnejad, A. Dick, I. Reid. In ICCV 2017
bibtex | abstract | paper | demo

@inproceedings{Rezatofighi:2017:ICCV,
	Title = {Deep{S}et{N}et: {P}redicting Sets with Deep Neural Networks},
	shorttitle = {DeepSetNet},
	url = {https://arxiv.org/abs/1611.08998},
	Author = {Rezatofighi, S. H. and Kumar BG, V. and Milan, A. and Abbasnejad, E. and Dick, A. and Reid, I.},
	Booktitle = {ICCV},
	Year = {2017}
}
			
This paper addresses the task of set prediction using deep learning. This is
important because the output of many computer vision tasks, including image
tagging and object detection, are naturally expressed as sets of entities rather
than vectors. As opposed to a vector, the size of a set is not fixed in advance,
and it is invariant to the ordering of entities within it. We define a
likelihood for a set distribution and learn its parameters using a deep neural
network. We also derive a loss for predicting a discrete distribution
corresponding to set cardinality. Set prediction is demonstrated on the problems
of multi-class image classification and pedestrian detection. Our approach
yields state-of-the-art results in both cases on standard datasets.
			

Multi-class RGB-D Object Detection and Semantic Segmentation for Autonomous Manipulation in Clutter

M. Schwarz, A. Milan, A. S. Periyasamy, S. Behnke.
Accepted to IJRR
bibtex | abstract

@article{Schwarz:2017:IJRR,
	title = {Multi-class RGB-D Object Detection and Semantic Segmentation for Autonomous Manipulation in Clutter},
	booktitle = {IJRR},
	author = {Schwarz, M. and Milan, A. and Periyasamy A. S. and Behnke S.},
  volume = {37},
  number = {4-5},
	month = {Jun},
	year = {2017}
}
			
Autonomous robotic manipulation in clutter is challenging. A large variety of objects must be perceived in complex scenes, where they are partially occluded and embedded among many distractors, often in restricted spaces. To tackle these challenges, we developed a deep-learning approach that combines object detection and semantic segmentation. The manipulation scenes are captured with RGB-D cameras, for which we developed a depth fusion method. Employing pretrained features makes learning from small annotated robotic data sets possible. We evaluate our approach on two challenging data sets: one captured for the Amazon Picking Challenge 2016, where our team NimbRo came in second in the Stowing and third in the Picking task, and one captured in disaster-response scenarios. The experiments show that object detection and semantic segmentation complement each other and can be combined to yield reliable object perception.
			

Tracking the Trackers:
An Analysis of the State of the Art in Multiple Object Tracking

L. Leal-Taixé, A. Milan, K. Schindler, D. Cremers, I. Reid, S. Roth
arXiv:1704.02781
bibtex | abstract | paper

@article{2017:Leal:arxiv,
	title = {Tracking the Trackers: {A}n Analysis of the State of the Art in Multiple Object Tracking},
	shorttitle = {Tracking the Trackers},
	url = {http://arxiv.org/abs/1704.02781},
	journal = {arXiv:1704.02781 [cs]},
	author = {Leal-Taix\'{e}, Laura and Milan, Anton and Schindler, Konrad and Cremers, Daniel and Reid, Ian and Roth, Stefan},
	month = apr,
	year = {2017},
	note = {arXiv: 1704.02781},
	keywords = {Computer Science - Computer Vision and Pattern Recognition}
}
			
Standardized benchmarks are crucial for the majority of
computer vision applications. Although leaderboards and
ranking tables should not be over-claimed, benchmarks often
provide the most objective measure of performance and
are therefore important guides for research. We present a
benchmark for Multiple Object Tracking launched in the
late 2014, with the goal of creating a framework for the
standardized evaluation of multiple object tracking methods.
This paper collects the two releases of the benchmark
made so far, and provides an in-depth analysis of almost
50 state-of-the-art trackers that were tested on over 11000
frames. We show the current trends and weaknesses of multiple
people tracking methods, and provide pointers of what
researchers should be focusing on to push the field forward.
			

RefineNet: Multi-Path Refinement Networks
for High-Resolution Semantic Segmentation

G. Lin, A. Milan, C. Shen, I. Reid. In CVPR 2017
bibtex | abstract | paper | code

@inproceedings{Lin:2017:CVPR,
        Author = {Lin, G. and Milan, A. and Shen, C. and Reid, I.},
	Title = {Refine{N}et: {M}ulti-Path Refinement Networks for High-Resolution Semantic Segmentation},
	Booktitle = {CVPR},
	Year = {2017}
}
			
Recently, very deep convolutional neural networks (CNNs) have shown
outstanding performance in object recognition and have also been the
first choice for dense classification problems such as semantic
segmentation. However, repeated subsampling operations like pooling or
convolution striding in deep CNNs lead to a significant decrease in the
initial image resolution. Here, we present RefineNet, a generic
multi-path refinement network that explicitly exploits all the
information available along the down-sampling process to enable
high-resolution prediction using long-range residual connections. In
this way, the deeper layers that capture high-level semantic features
can be directly refined using fine-grained features from earlier
convolutions. The individual components of RefineNet employ residual
connections following the identity mapping mindset, which allows for
effective end-to-end training. Further, we introduce chained residual
pooling, which captures rich background context in an efficient manner.
We carry out comprehensive experiments and set new state-of-the-art
results on seven public datasets. In particular, we achieve an
intersection-over-union score of 83.4 on the challenging PASCAL VOC 2012
dataset, which is the best reported result to date.
			

PoseTrack: Joint Multi-Person Pose Estimation and Tracking

U. Iqbal, A. Milan, J. Gall. In CVPR 2017
bibtex | abstract | paper | project

@inproceedings{Iqbal:2017:CVPR,
        Author = {Iqbal, U. and Milan, A. and Gall, J.},
	Title = {Pose{T}rack: {J}oint Multi-Person Pose Estimation and Tracking},
	Booktitle = {CVPR},
	Year = {2017}
}
			
In this work, we introduce the challenging problem of joint multi-person pose
estimation and tracking of an unknown number of persons in unconstrained videos.
Existing methods for multi-person pose estimation in images cannot be applied
directly to this problem, since it also requires to solve the problem of person
association over time in addition to the pose estimation for each person. We
therefore propose a novel method that jointly models multi-person pose
estimation and tracking in a single formulation. To this end, we represent body
joint detections in a video by a spatio-temporal graph and solve an integer
linear program to partition the graph into sub-graphs that correspond to
plausible body pose trajectories for each person. The proposed approach
implicitly handles occlusions and truncations of persons. Since the problem has
not been addressed quantitatively in the literature, we introduce a challenging
"Multi-Person Pose-Track" dataset, and also propose a completely unconstrained
evaluation protocol that does not make any assumptions on the scale, size,
location or the number of persons. Finally, we evaluate the proposed approach
and several baseline methods on our new dataset.
			

NimbRo Picking: Versatile Part Handling for Warehouse Automation

M. Schwarz, A. Milan, C. Lenz, A. Muñoz, A. S. Periyasamy, M. Schreiber, S. Schüller, S. Behnke.
In ICRA 2017
bibtex | abstract | paper

@inproceedings{Schwarz:2017:ICRA,
	title = {Nimb{R}o {P}icking: {V}ersatile Part Handling for Warehouse Automation},
	booktitle = {ICRA},
	author = {Schwarz, M. and Milan, A. and Lenz C. and Mu{\~n}oz A. and Periyasamy A. S. and Schreiber M. and Sch{\"u}ller S. and Behnke S.},
	month = {June},
	year = {2017}
}
			
Part handling in warehouse automation is challenging if a large variety of items
must be accommodated and items are stored in unordered piles. To foster research
in this domain, Amazon holds picking challenges. We present our system which
achieved second and third place in the Amazon Picking Challenge 2016 tasks. The
challenge required participants to pick a list of items from a shelf or to stow
items into the shelf. Using two deep-learning approaches for object detection
and semantic segmentation and one item model registration method, our system
localizes the requested item. Manipulation occurs using suction on points
determined heuristically or from 6D item model registration. Parametrized motion
primitives are chained to generate motions. We present a full-system evaluation
during the APC 2016 and component level evaluations of the perception system on
an annotated dataset.
			

Data-Driven Approximations to NP-Hard Problems

A. Milan, S. H. Rezatofighi, R. Garg, A. Dick, I. Reid. In AAAI 2017
bibtex | abstract | paper | slides | code

@inproceedings{Milan:2017:AAAI_NP,
	title = {Data-driven approximations to {NP}-hard problems},
	booktitle = {AAAI},
	author = {Milan, A. and Rezatofighi, S. H. and Garg, R. and Dick, A. and Reid, I.},
	month = {February},
	year = {2017}
}
			
There exist a number of problem classes, for which obtaining
the exact solution becomes exponentially expensive with
increasing problem size.  The quadratic assignment problem (QAP)
or the travelling salesman problem (TSP) are just two examples of
such NP-hard problems. In practice, approximate algorithms
are employed to obtain a suboptimal solution, where one must
face a trade-off between computational complexity and
solution quality.
In this paper, we propose to learn  to solve these problem from
approximate examples, using recurrent neural networks (RNNs).
Surprisingly, such architectures are capable of producing
highly accurate solutions at minimal computational cost.
Moreover, we introduce a simple,
yet effective technique for improving the initial (weak) training
set by incorporating the objective cost into the training procedure.
We demonstrate the functionality of our approach on three
exemplar applications: marginal distributions of a joint
matching space, feature point matching and the travelling
salesman problem. We show encouraging results on synthetic and real data
in all three cases.
			

Online Multi-Target Tracking Using Recurrent Neural Networks

A. Milan, S. H. Rezatofighi, A. Dick, I. Reid, K. Schindler
In AAAI 2017
bibtex | abstract | paper | poster | code

@inproceedings{Milan:2017:AAAI_RNNTracking,
	title = {Online Multi-Target Tracking using Recurrent Neural Networks},
	booktitle = {AAAI},
	author = {Milan, A. and Rezatofighi, S. H. and Dick, A. and Reid, I. and Schindler, K.},
	month = {February},
	year = {2017}
}
			
We present a novel approach to online multi-target tracking based on
recurrent neural networks (RNNs). Tracking multiple objects in
real-world scenes involves many challenges, including a) an a-priori
unknown and time-varying number of targets, b) a continuous state
estimation of all present targets, and c) a discrete combinatorial
problem of data association. Most previous methods involve complex
models that require tedious tuning of parameters. Here, we propose for
the first time, a full end-to-end learning approach for online
multi-target tracking based on deep learning. Existing deep learning
methods are not designed for the above challenges and cannot be
trivially applied to the task. Our solution addresses all of the above
points in a principled way. Experiments on both synthetic and real data
show competitive results obtained at 300 Hz on a standard CPU, and pave
the way towards future research in this direction.
			


2016

Joint Probabilistic Matching Using m-Best Solutions

S. H. Rezatofighi, A. Milan, Z. Zhang, A. Dick, Q. Shi, I. Reid
CVPR 2016 (oral presentation)
bibtex | abstract | paper | supplemental | slides | poster | video | project

@inproceedings{Rezatofighi:2016:CVPR,
	Author = {Rezatofighi, S. H. and Milan, A. and Zhang, Z. and Shi, Q. and Dick, A. and Reid, I.},
	Booktitle = {CVPR},
	Title = {Joint Probabilistic Matching Using m-Best Solutions},
	Year = {2016}
}
			
Matching between two sets of objects is typically ap-
proached by finding the object pairs that collectively maxi-
mize the joint matching score. In this paper, we argue that
this single solution does not necessarily lead to the opti-
mal matching accuracy and that general one-to-one assign-
ment problems can be improved by considering multiple hy-
potheses before computing the final similarity measure. To
that end, we propose to utilize the marginal distributions for
each entity. Previously, this idea has been neglected mainly
because exact marginalization is intractable due to a com-
binatorial number of all possible matching permutations.
Here, we propose a generic approach to efficiently approx-
imate the marginal distributions by exploiting the m-best
solutions of the original problem. This approach not only
improves the matching solution, but also provides more ac-
curate ranking of the results, because of the extra informa-
tion included in the marginal distribution. We validate our
claim on two distinct objectives: (i) person re-identification
and temporal matching modeled as an integer linear pro-
gram, and (ii) feature point matching using a quadratic cost
function. Our experiments confirm that marginalization in-
deed leads to superior performance compared to the single
(nearly) optimal solution, yielding state-of-the-art
results in both applications on standard benchmarks.

			

MOT16: A Benchmark for Multi-Object Tracking

A. Milan, L. Leal-Taixé, I. Reid, S. Roth, and K. Schindler
arXiv:1603.00831
bibtex | abstract | paper | project page

@article{MOT16,
	title = {{MOT}16: {A} Benchmark for Multi-Object Tracking},
	shorttitle = {MOT16},
	url = {http://arxiv.org/abs/1603.00831},
	journal = {arXiv:1603.00831 [cs]},
	author = {Milan, A. and Leal-Taix\'{e}, L. and Reid, I. and Roth, S. and Schindler, K.},
	month = mar,
	year = {2016},
	note = {arXiv: 1603.00831},
	keywords = {Computer Science - Computer Vision and Pattern Recognition}
}
			
Standardized benchmarks are crucial for the majority of computer vision
applications. Although leaderboards and ranking tables should not be
over-claimed, benchmarks often provide the most objective measure of
performance and are therefore important guides for reseach. Recently, a
new benchmark for Multiple Object Tracking, MOTChallenge, was launched
with the goal of collecting existing and new data and creating a
framework for the standardized evaluation of multiple object tracking
methods. The first release of the benchmark focuses on multiple people
tracking, since pedestrians are by far the most studied object in the
tracking community. This paper accompanies a new release of the
MOTChallenge benchmark. Unlike the initial release, all videos of MOT16
have been carefully annotated following a consistent protocol. Moreover,
it not only offers a significant increase in the number of labeled
boxes, but also provides multiple object classes beside pedestrians and
the level of visibility for every single object of interest.


			

Multi-Target Tracking by Discrete-Continuous Energy Minimization

A. Milan, K. Schindler and S. Roth
PAMI 38(1), 2016
bibtex | abstract | paper | video | project page

@article{Milan:2016:PAMI,
	author = {Milan, A. and Schindler, K. and Roth, S.},
	title = {Multi-Target Tracking by Discrete-Continuous Energy Minimization},
	volume = {38},
	number = {10},
	month={Oct},
	pages={2054-2068},
	doi={10.1109/TPAMI.2015.2505309},
	journal = {IEEE TPAMI},
	year = {2016}
}
			
The task of tracking multiple targets is often addressed with the
so-called tracking-by-detection paradigm, where the first step is to
obtain a set of target hypotheses for each frame independently. Tracking
can then be regarded as solving two separate, but tightly coupled
problems. The first is to carry out data association, i.e. to determine
the origin of each of the available observations. The second problem is
to reconstruct the actual trajectories that describe the spatio-temporal
motion pattern of each individual target. The former is inherently a
discrete problem, while the latter should intuitively be modeled in
continuous space. Having to deal with an unknown number of targets,
complex dependencies, and physical constraints, both are challenging
tasks on their own and thus most previous work focuses on one of these
subproblems. Here, we present a multi-target tracking approach that
explicitly models both tasks as minimization of a unified
discrete-continuous energy function. Trajectory properties are captured
through global label costs, a recent concept from multi-model fitting,
which we introduce to tracking. Specifically, label costs describe
physical properties of individual tracks, e.g. linear and angular
dynamics, or entry and exit points. We further introduce pairwise label
costs to describe mutual interactions between targets in order to avoid
collisions. By choosing appropriate forms for the individual energy
components, powerful discrete optimization techniques can be leveraged
to address data association, while the shapes of individual trajectories
are updated by gradient-based continuous energy minimization. The
proposed method achieves state-of-the-art results on diverse benchmark
sequences.
			


2015

Joint Probabilistic Data Association Revisited

S. H. Rezatofighi, A. Milan, Z. Zhang, A. Dick, Q. Shi, I. Reid
ICCV 2015
bibtex | abstract | paper | code | video 1 video 2

@inproceedings{Rezatofighi:2015:ICCV,
	Author = {Rezatofighi, S. H. and Milan, A. and Zhang, Z. and Shi, Q. and Dick, A. and Reid, I.},
	Booktitle = {ICCV},
	Title = {Joint Probabilistic Data Association Revisited},
	Year = {2015}
}
			
In this paper, we revisit the joint probabilistic data association (JPDA) technique and propose a novel solution based on recent developments in finding the m-best solutions to an integer linear program. The key advantage of this approach is that it makes JPDA computationally tractable in applications with high target and/or clutter density, such as spot tracking in fluorescence microscopy sequences and pedestrian tracking in surveillance footage. We also show that our JPDA algorithm embedded in a simple tracking framework is surprisingly competitive with state-of-the-art global tracking methods in these two applications, while needing considerably less processing time.
			

MOTChallenge 2015: Towards a Benchmark for Multi-Target Tracking

L. Leal-Taixé, A. Milan, I. Reid, S. Roth, and K. Schindler
arXiv:1504.01942
bibtex | abstract | paper | project page

@article{MOTChallenge2015,
	title = {MOTChallenge 2015: Towards a Benchmark for Multi-Target Tracking},
	shorttitle = {MOTChallenge 2015},
	url = {http://arxiv.org/abs/1504.01942},
	journal = {arXiv:1504.01942 [cs]},
	author = {Leal-Taix\'{e}, Laura and Milan, Anton and Reid, Ian and Roth, Stefan and Schindler, Konrad},
	month = apr,
	year = {2015},
	note = {arXiv: 1504.01942},
	keywords = {Computer Science - Computer Vision and Pattern Recognition}
}
			
In the recent past, the computer vision community has developed
centralized benchmarks for the performance evaluation of a variety of
tasks, including generic object and pedestrian detection, 3D
reconstruction, optical flow, single-object short-term tracking, and
stereo estimation. Despite potential pitfalls of such benchmarks, they
have proved to be extremely helpful to advance the state of the art in
the respective area. Interestingly, there has been rather limited work
on the standardization of quantitative benchmarks for multiple target
tracking. One of the few exceptions is the well-known PETS dataset,
targeted primarily at surveillance applications. Despite being widely
used, it is often applied inconsistently, for example involving using
different subsets of the available data, different ways of training the
models, or differing evaluation scripts. This paper describes our work
toward a novel multiple object tracking benchmark aimed to address such
issues. We discuss the challenges of creating such a framework,
collecting existing and new data, gathering state-of-the-art methods to
be tested on the datasets, and finally creating a unified evaluation
system. With MOTChallenge we aim to pave the way toward a unified
evaluation framework for a more meaningful quantification of
multi-target tracking.


			

Joint Tracking and Segmentation of Multiple Targets

A. Milan, L. Leal-Taixé, K. Schindler and I. Reid
CVPR 2015
bibtex | abstract | ext. abstract | paper | video | project page

@inproceedings{Milan:2015:CVPR,
	Author = {Anton Milan and Laura Leal-Taixé and Konrad Schindler and Ian Reid},
	Booktitle = {CVPR},
	Title = {Joint Tracking and Segmentation of Multiple Targets},
	Year = {2015}
}
			
Tracking-by-detection has proven to be the most successful strategy to
address the task of tracking multiple targets in unconstrained
scenarios. Traditionally, a set of sparse detections, generated in a
preprocessing step, serves as input to a high-level tracker whose goal
is to correctly associate these “dots” over time. An obvious shortcoming
of this approach is that most information available in image sequences
is simply ignored by thresholding weak detection responses and applying
non-maximum suppression. We propose a multi-target tracker that exploits
low level image information and associates every (super)-pixel to a
specific target or classifies it as background. As a result, we obtain a
video segmentation in addition to the classical bounding-box
representation in unconstrained, realworld videos. Our method shows
encouraging results on many standard benchmark sequences and
significantly outperforms state-of-the-art tracking-by-detection
approaches in crowded scenes with long-term partial occlusions.
			


2014

Privacy Preserving Multi-target Tracking

A. Milan, S. Roth, K. Schindler and M. Kudo
Workshop on Human Identification for Surveillance (HIS)
bibtex | abstract | paper | video | slides | project page

@INPROCEEDINGS{Milan:2014:ACCVWS,
  author = {A.Milan and S. Roth and K. Schindler and M. Kudo},
  title = {Privacy Preserving Multi-target Tracking},
  booktitle = {Workshop on Human Identification for Surveillance},
  year = {2014}
}
			
Automated people tracking is important for a wide range
of applications. However, typical surveillance cameras are controversial
in their use, mainly due to the harsh intrusion of the tracked individ-
uals’ privacy. In this paper, we explore a privacy-preserving alternative
for multi-target tracking. A network of infrared sensors attached to the
ceiling acts as a low-resolution, monochromatic camera in an indoor en-
vironment. Using only this low-level information about the presence of
a target, we are able to reconstruct entire trajectories of several peo-
ple. Inspired by the recent success of offline approaches to multi-target
tracking [1–3], we apply an energy minimization technique to the novel
setting of infrared motion sensors. To cope with the very weak data term
from the infrared sensor network we track in a continuous state space
with soft, implicit data association. Our experimental evaluation on both
synthetic and real-world data shows that our principled method clearly
outperforms previous techniques.
			


Improving Global Multi-target Tracking with Local Updates

A. Milan, R. Gade, A. Dick, T. B. Moeslund, I. Reid
Workshop on Visual Surveillance and Re-Identification
bibtex | abstract | paper | video | slides

@INPROCEEDINGS{Milan:2014:IGM,
  author = {A. Milan and R. Gade and A. Dick and T. B. Moeslund and I. Reid},
  title = {Improving Global Multi-target Tracking with Local Updates},
  booktitle = {Workshop on Visual Surveillance and Re-Identification},
  year = {2014}
}
			
We propose a scheme to explicitly detect and resolve ambiguous
situations in multiple target tracking. During periods of uncertainty,
our method applies multiple local single target trackers to hypothesise
short term tracks. These tracks are combined with the tracks obtained
by a global multi-target tracker, if they result in a reduction in the global
cost function. Since tracking failures typically arise when targets become
occluded, we propose a local data association scheme to maintain the
target identities in these situations. We demonstrate a reduction of up
to 50% in the global cost function, which in turn leads to superior
performance on several challenging benchmark sequences. Additionally, we
show tracking result in sports videos where poor video quality and
frequent and severe occlusion
			


Energy Minimization for Multiple Object Tracking

A. Milan
PhD Thesis
bibtex | thesis | slides |

@phdthesis{Milan:2014:EMM,
	address = {Darmstadt},
	type = {{PhD}},
	title = {Energy Minimization for Multiple Object Tracking},
	url = {http://tuprints.ulb.tu-darmstadt.de/3463/},
	school = {{TU} Darmstadt},
	author = {Milan, Anton},
	year = {2014},
}
			


Continuous Energy Minimization for Multitarget Tracking

A. Milan, S. Roth and K. Schindler
PAMI 36(1), 2014
bibtex | abstract | paper | video | slides | project page

@article{Milan:2014:CEM,
	author = {Milan, A. and Roth, S. and Schindler, K.},
	title = {Continuous Energy Minimization for Multitarget Tracking},
	volume = {36},
	issn = {0162-8828},
	doi = {10.1109/TPAMI.2013.103},
	number = {1},
	journal = {IEEE TPAMI},
	year = {2014},
	pages = {58--72}
}
			
Many recent advances in multiple target tracking aim at finding a
(nearly) optimal set of trajectories within a temporal window. To handle
the large space of possible trajectory hypotheses, it is typically
reduced to a finite set by some form of data-driven or regular
discretization. In this work we propose an alternative formulation of
multi-target tracking as minimization of a continuous energy. Contrary
to recent approaches, we focus on designing an energy that corresponds
to a more complete representation of the problem, rather than one that
is amenable to global optimization. Besides the image evidence, the
energy function takes into account physical constraints, such as target
dynamics, mutual exclusion, and track persistence. In addition, partial
image evidence is handled with explicit occlusion reasoning, and
different targets are disambiguated with an appearance model. To
nevertheless find strong local minima of the proposed non-convex energy
we construct a suitable optimization scheme that alternates between
continuous conjugate gradient descent and discrete trans-dimensional
jump moves. These moves, which are executed such that they always reduce
the energy, allow the search to escape weak minima and explore a much
larger portion of the search space of varying dimensionality. We
demonstrate the validity of our approach with an extensive quantitative
evaluation on several public datasets.
			


2013

Learning People Detectors for Tracking in Crowded Scenes

S. Tang, M. Andriluka, A. Milan, K. Schindler, S. Roth and B. Schiele
ICCV 2013
bibtex | abstract | paper | poster | video | project page

@INPROCEEDINGS{Tang:2013:LPD,
  author = {S. Tang and M. Andriluka and A. Milan and K. Schindler and S. Roth and B. Schiele},
  title = {Learning People Detectors for Tracking in Crowded Scenes},
  booktitle = {ICCV},
  year = {2013}
}
			
People tracking in crowded real-world scenes is challenging due to
frequent and long-term occlusions. Recent tracking methods obtain the
image evidence from object (people) detectors, but typically use
off-the-shelf detectors and treat them as black box components. In this
paper we argue that for best performance one should explicitly train
people detectors on failure cases of the overall tracker instead. To
that end, we first propose a novel joint people detector that combines a
state-of-the-art single person detector with a detector for pairs of
people, which explicitly exploits common patterns of person-person
occlusions across multiple viewpoints that are a common failure case for
tracking in crowded scenes. To explicitly address remaining failure
cases of the tracker we explore two methods. First, we analyze typical
failure cases of trackers and train a detector explicitly on those
failure cases. And second, we train the detector with the people tracker
in the loop, focusing on the most common tracker failures. We show that
our joint multi-person detector significantly improves both detection
accuracy as well as tracker performance, improving the state-of-the-art
on standard benchmarks.
			


Challenges of Ground Truth Evaluation of Multi-Target Tracking

A. Milan, K. Schindler and S. Roth
CVPR Workshop on Ground Truth
bibtex | abstract | paper | poster

@inproceedings{Milan:2013:CGT,
	Author = {Anton Milan and Konrad Schindler and Stefan Roth},
	Booktitle = {Proc. of the CVPR 2013 Workshop on Ground Truth - What is a good dataset?},
	Title = {Challenges of Ground Truth Evaluation of Multi-Target Tracking},
	Year = {2013}
}
			
Evaluating multi-target tracking based on ground truth data is a
surprisingly challenging task. Erroneous or ambiguous ground truth
annotations, numerous evaluation protocols, and the lack of standardized
benchmarks make a direct quantitative comparison of different tracking
approaches rather difficult. The goal of this paper is to raise
awareness of common pitfalls related to objective ground truth
evaluation. We investigate the influence of different annotations,
evaluation software, and training procedures using several publicly
available resources, and point out the limitations of current
definitions of evaluation metrics. Finally, we argue that the
development an extensive standardized benchmark for multi-target
tracking is an essential step toward more objective comparison of
tracking approaches.
			


Detection- and Trajectory-Level Exclusion in Multiple Object Tracking

A. Milan, K. Schindler and S. Roth
CVPR 2013
bibtex | abstract | paper | poster | video | slides | data | project page

@inproceedings{Milan:2013:DTE,
	Author = {Anton Milan and Konrad Schindler and Stefan Roth},
	Booktitle = {CVPR},
	Title = {Detection- and Trajectory-Level Exclusion in Multiple Object Tracking},
	Year = {2013}
}
			
When tracking multiple targets in crowded scenarios, modeling mutual
exclusion between distinct targets becomes important at two levels: (1)
in data association, each target observation should support at most one
trajectory and each trajectory should be assigned at most one
observation per frame; (2) in trajectory estimation, two trajectories
should remain spatially separated at all times to avoid collisions. Yet,
existing trackers often sidestep these important constraints. We address
this using a mixed discrete-continuous conditional random field (CRF)
that explicitly models both types of constraints: Exclusion between
conflicting observations with supermodular pairwise terms, and exclusion
between trajectories by generalizing global label costs to suppress the
co-occurrence of incompatible labels (trajectories). We develop an
expansion move-based MAP estimation scheme that handles both
non-submodular constraints and pairwise global label costs. Furthermore,
we perform a statistical analysis of ground-truth trajectories to derive
appropriate CRF potentials for modeling data fidelity, target dynamics,
and inter-target occlusion.
			


2012

Discrete-Continuous Optimization for Multi-Target Tracking

A. Andriyenko, K. Schindler and S. Roth
CVPR 2012
bibtex | abstract | paper | poster | video | data | project page

@inproceedings{Andriyenko:2012:DCO,
	Author = {Anton Andriyenko and Konrad Schindler and Stefan Roth},
	Booktitle = {CVPR},
	Title = {Discrete-Continuous Optimization for Multi-Target Tracking},
	Year = {2012}
}
			
The problem of multi-target tracking is comprised of two distinct, but
tightly coupled challenges: (i) the naturally discrete problem of data
association, i.e. assigning image observations to the appropriate
target; (ii) the naturally continuous problem of trajectory estimation,
i.e. recovering the trajectories of all targets. To go beyond simple
greedy solutions for data association, recent approaches often perform
multi-target tracking using discrete optimization. This has the
disadvantage that trajectories need to be pre-computed or represented
discretely, thus limiting accuracy. In this paper we instead formulate
multi-target tracking as a discretecontinuous optimization problem that
handles each aspect in its natural domain and allows leveraging powerful
methods for multi-model fitting. Data association is performed using
discrete optimization with label costs, yielding near optimality.
Trajectory estimation is posed as a continuous fitting problem with a
simple closed-form solution, which is used in turn to update the label
costs. We demonstrate the accuracy and robustness of our approach with
state-of-theart performance on several standard datasets.
			


2011

An Analytical Formulation of Global Occlusion Reasoning for Multi-Target Tracking

A. Andriyenko, S. Roth and K. Schindler
ICCV Workshop on Visual Surveillance
bibtex | abstract | paper | poster | video | slides | project page

@inproceedings{Andriyenko:2011:AFG,
	Author = {Anton Andriyenko and Stefan Roth and Konrad Schindler},
	Booktitle = {Proc. of the 11th International IEEE Workshop on Visual Surveillance},
	Title = {An Analytical Formulation of Global Occlusion Reasoning for Multi-Target Tracking},
	Year = {2011}
}
			
We present a principled model for occlusion reasoning in complex
scenarios with frequent inter-object occlusions, and its application to
multi-target tracking. To compute the putative overlap between pairs of
targets, we represent each target with a Gaussian. Conveniently, this
leads to an analytical form for the relative overlap – another Gaussian
– which is combined with a sigmoidal term for modeling depth relations.
Our global occlusion model bears several advantages: Global target
visibility can be computed efficiently in closed-form, and varying
degrees of partial occlusion can be naturally accounted for. Moreover,
the dependence of the occlusion on the target locations – i.e. the
gradient of the overlap – can also be computed in closedform, which
makes it possible to efficiently include the proposed occlusion model in
a continuous energy minimization framework. Experimental results on
seven datasets confirm that the proposed formulation consistently
reduces missed targets and lost trajectories, especially in challenging
scenarios with crowds and severe inter-object occlusions.
			


Multi-target Tracking by Continuous Energy Minimization

A. Andriyenko and K. Schindler
CVPR 2011
bibtex | abstract | paper | poster | video | slides | project page

@inproceedings{Andriyenko:2011:MTT,
	Author = {Anton Andriyenko and Konrad Schindler},
	Booktitle = {CVPR},
	Title = {Multi-target Tracking by Continuous Energy Minimization},
	Year = {2011}
}
			
We propose to formulate multi-target tracking as minimization of a
continuous energy function. Other than a number of recent approaches we
focus on designing an energy function that represents the problem as
faithfully as possible, rather than one that is amenable to elegant
optimization. We then go on to construct a suitable optimization scheme
to find strong local minima of the proposed energy. The scheme extends
the conjugate gradient method with periodic trans-dimensional jumps.
These moves allow the search to escape weak minima and explore a much
larger portion of the variable-dimensional search space, while still
always reducing the energy. To demonstrate the validity of this approach
we present an extensive quantitative evaluation both on synthetic data
and on six different real video sequences. In both cases we achieve a
significant performance improvement over an extended Kalman filter
baseline as well as an ILP-based state-of-the-art tracker.
			


2010

Globally Optimal Multi-target Tracking on a Hexagonal Lattice

A. Andriyenko and K. Schindler
ECCV 2010
bibtex | abstract | poster | video

@inproceedings{Andriyenko:2010:GOM,
	Author = {Anton Andriyenko and Konrad Schindler},
	Booktitle = {ECCV},
	Title = {Globally Optimal Multi-target Tracking on a Hexagonal Lattice},
	Year = {2010}
}
			
We propose a global optimisation approach to multi-target tracking. The
method extends recent work which casts tracking as an integer linear
program, by discretising the space of target locations. Our main
contribution is to show how dynamic models can be integrated in such an
approach. The dynamic model, which encodes prior expectations about
object motion, has been an important component of tracking systems for a
long time, but has recently been dropped to achieve globally optimisable
objective functions. We re-introduce it by formulating the optimisation
problem such that deviations from the prior can be measured
independently for each variable. Furthermore, we propose to sample the
location space on a hexagonal lattice to achieve smoother, more accurate
trajectories in spite of the discrete setting. Finally, we argue that
non-maxima suppression in the measured evidence should be performed
during tracking, when the temporal context and the motion prior are
available, rather than as a preprocessing step on a per-frame basis.
Experiments on five different recent benchmark sequences demonstrate the
validity of our approach.
			


2009

A Practical Approach for Photometric Acquisition of Hair Color

A. Zinke, M. Rump, T. Lay, A. Weber, A. Andriyenko and R. Klein
SIGGRAPH ASIA 2009
bibtex | abstract

@inproceedings{Zinke:2009:PAP,
	Author = {Zinke, Arno and Rump, Martin and Lay, Tom\'{a}s and Weber, Andreas and Andriyenko, Anton and Klein, Reinhard},
	Booktitle = {ACM SIGGRAPH Asia 2009 papers},
	Title = {A Practical Approach for Photometric Acquisition of Hair Color},
	Year = {2009}
}
			
In this work a practical approach to photometric acquisition of hair
color is presented. Based on a single input photograph of a simple setup
we are able to extract physically plausible optical properties of hair
and to render virtual hair closely matching the original. Our approach
does not require any costly special hardware but a standard consumer
camera only.
			


Code


Our code is available for the following papers

Data


Annotations, detections, and other data for selected datasets are available here.

Contact



Amazon Development Center
Krausenstr. 38
10117 Berlin, Germany
E-Mail: antmila@amazon (replace amazon with amazon.com)