Sara Sabour
Research Areas
Authored Publications
Google Publications
Other Publications
Sort By
Conditional Object-Centric Learning from Video
Thomas Kipf
Austin Stone
Rico Jonschkowski
Alexey Dosovitskiy
Klaus Greff
ICLR, ICLR (2022)
Preview abstract
Object-centric representations are a promising path toward more systematic generalization by providing flexible abstractions upon which compositional world models can be built. Recent work on simple 2D and 3D datasets has shown that models with object-centric inductive biases can learn to segment and represent meaningful objects from the statistical structure of the data alone without the need for any supervision. However, such fully-unsupervised methods still fail to scale to diverse realistic data, despite the use of increasingly complex inductive biases such as priors for the size of objects or the 3D geometry of the scene. In this paper, we instead take a weakly-supervised approach and focus on how 1) using the temporal dynamics of video data in the form of optical flow and 2) conditioning the model on simple object location cues can be used to enable segmenting and tracking objects in significantly more realistic synthetic data. We introduce a sequential extension to Slot Attention which we train to predict optical flow for realistic looking synthetic scenes and show that conditioning the initial state of this model on a small set of hints, such as center of mass of objects in the first frame, is sufficient to significantly improve instance segmentation. These benefits generalize beyond the training distribution to novel objects, novel backgrounds, and to longer video sequences. We also find that such initial-state-conditioning can be used during inference as a flexible interface to query the model for specific objects or parts of objects, which could pave the way for a range of weakly-supervised approaches and allow more effective interaction with trained models.
View details
Kubric: A scalable dataset generator
Anissa Yuenming Mak
Austin Stone
Carl Doersch
Cengiz Oztireli
Charles Herrmann
Daniel Rebain
Derek Nowrouzezahrai
Dmitry Lagun
Fangcheng Zhong
Florian Golemo
Francois Belletti
Henning Meyer
Hsueh-Ti (Derek) Liu
Issam Laradji
Klaus Greff
Kwang Moo Yi
Matan Sela
Noha Radwan
Thomas Kipf
Tianhao Wu
Vincent Sitzmann
Yilun Du
Yishu Miao
(2022)
Preview abstract
Data is the driving force of machine learning. The amount and quality of training data is often more important for the performance of a system than the details of its architecture. Data is also an important tool for testing specific hypothesis, and for empirically evaluating the behaviour of complex systems. Synthetic data generation represents a powerful tool that can address all these shortcomings: 1) it is cheap 2) supports rich ground-truth annotations 3) offers full control over data and 4) can circumvent privacy and legal concerns. Unfortunately the toolchain for generating data is less well developed than that for building models. We aim to improve this situation by introducing Kubric: a scalable open-source pipeline for generating realistic image and video data with rich ground truth annotations.
We also publish a collection of generated datasets and baseline results on several vision tasks.
View details
Canonical Capsules: Unsupervised Capsules in Canonical Pose
Weiwei Sun
Boyang Deng
Soroosh Yazdani
Geoffrey Everest Hinton
Kwang Moo Yi
Neural Information Processing Systems (NeurIPS) (2021)
Preview abstract
We propose an unsupervised capsule architecture for 3D point clouds. We compute capsule decompositions of objects through permutation-equivariant attention, and self-supervise the process by training with pairs of randomly rotated objects. Our key idea is to aggregate the attention masks into semantic keypoints, and use these to supervise a decomposition that satisfies the capsule invariance/equivariance properties. This not only enables the training of a semantically consistent decomposition, but also allows us to learn a canonicalization operation that enables object-centric reasoning. In doing so, we require neither classification labels nor manually-aligned training datasets to train. Yet, by learning an object-centric representation in an unsupervised manner, our method outperforms the state-of-the-art on 3D point cloud reconstruction, registration, and unsupervised classification. We will release the code and dataset to reproduce our results as soon as the paper is published.
View details
Preview abstract
We present Optimal Completion Distillation (OCD), a training procedure for
optimizing sequence to sequence models based on edit distance. OCD is efficient,
has no hyper-parameters of its own, and does not require pretraining or joint
optimization with conditional log-likelihood. Given a partial sequence generated
by the model, we first identify the set of optimal suffixes that minimize the total
edit distance, using an efficient dynamic programming algorithm. Then, for each
position of the generated sequence, we define a target distribution that puts an equal
probability on the first token of each optimal suffix. OCD achieves the state-of-theart performance on end-to-end speech recognition, on both Wall Street Journal and
Librispeech datasets, achieving 9.3% and 4.5% word error rates, respectively.
View details
Preview abstract
An object can be seen as a geometrically organized set of interrelated parts. A system that makes explicit use of these geometric relationships to recognize objects should be naturally robust to changes in viewpoint, because the intrinsic geometric relationships are viewpoint-invariant. We describe an unsupervised version of capsule networks, in which a neural encoder, which looks at all of the parts, is used to infer the presence and poses of object capsules. The encoder is trained by backpropagating through a decoder, which predicts the pose of each already discovered part using a mixture of pose predictions. The parts are discovered directly from an image, in a similar manner, by using a neural encoder, which infers parts and their affine transformations. The corresponding decoder models each image pixel as a mixture of predictions made by affine-transformed parts. We learn object- and their part-capsules on unlabeled data, and then cluster the vectors of presences of object capsules. When told the names of these clusters, we achieve state-of-the-art results for unsupervised classification on SVHN (55%) and near state-of-the-art on MNIST (98.5%).
View details
Preview abstract
A capsule is a group of neurons whose outputs represent different properties of the same entity. We describe a version of capsules in which each capsule has a logistic unit to represent the presence of an entity and a 4x4 pose matrix to represent the relationship between that entity and the viewer. A capsule in one layer votes for the pose matrices of many different capsules in the layer above by multiplying its own pose matrix by viewpoint-invariant transformation matrices that represent part-whole relationships. Each of these votes is weighted by an assignment coefficient and these coefficients are iteratively updated using the EM algorithm so that the output of each capsule is routed to a capsule in the layer above that receives a cluster of similar votes. The whole system is trained discriminatively by unrolling the 3 iterations of EM between each pair of adjacent layers. On the small NORB benchmark, capsules reduce the number of test errors by 30\% compared with the best reported CNN. Capsules are also far more resistant to whitebox adversarial attack.
View details
Preview abstract
A capsule is a group of neurons whose activity vector represents the instantiation parameters of a specific type of entity such as an object or object part. We use the length of the activity vector to represent the probability that the entity exists and its orientation to represent the instantiation paramters. Active capsules at one level make predictions, via transformation matrices, for the instantiation parameters of higher-level capsules. When multiple predictions agree, a higher level capsule becomes active. We show that a discrimininatively trained, multi-layer capsule system achieves state-of-the-art performance on MNIST and is considerably better than a convolutional net at recognizing highly overlapping digits. To achieve these results we use an iterative routing-by-agreement mechanism: A lower-level capsule prefers to send its output to higher level capsules whose activity vectors have a big scalar product with the prediction coming from the lower-level capsule. The final version of the paper is under revision to encorporate reviewers comments.
View details
No Results Found
