May 24, 2023 · Here we introduce architectural modifications and a novel contrastive learning method that greatly improve the state-of-the-art synchrony-based ...

All these results demonstrate that CtCAE is capable of object discovery (still far from perfect) on all datasets which include color images and more than three ...

This repo provides a reference implementation for the CtCAE as introduced in our paper "Contrastive Training of Complex-Valued Autoencoders for Object Discovery ...

May 30, 2024 · Current state-of-the-art object-centric models use slots and attention-based routing for binding. However, this class of models has several ...

May 24, 2023 · The discovery of reusable subroutines simplifies decision making and planning in complex reinforcement learning problems. Previous approaches ...

This work proposes a comparatively simple approach - the Complex AutoEncoder (CAE) - that creates distributed object-centric representations that achieves ...

Alternative unsupervised training methods for SlotAttention with. e.g. contrastive learning remain restricted to simplistic environments (Löwe et al., 2020).

In our experiments, we show that this approach is slow to train due to its greedy layerwise training with Contrastive Divergence (Hinton, 2012), that it is slow ...

Missing: Contrastive | Show results with:Contrastive