Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

QASA: Quality-Guided K-Adaptive Slot Attention for Unsupervised Object-Centric Learning

Tianran Ouyang, Xingping Dong, Jing Zhang, Mang Ye, Jun Chen, Bo Du

TL;DR

This work addresses the challenge of varying object counts in unsupervised object-centric learning by introducing Quality-Guided K-Adaptive Slot Attention (QASA). It defines a Slot-Quality metric $Q_i$ and a greedy, coverage-driven Quality-Guided Slot Selection to pick high-quality slots for reconstruction, while decoupling slot selection from reconstruction with a gated decoder. The approach supports both Transformer and MLP decoders and yields state-of-the-art results on real-world data (e.g., COCO, VOC) and competitive performance on synthetic data, outperforming existing $K$-adaptive baselines by up to $8.4\%$ in mean best overlap and surpassing many $K$-fixed methods. Ablation studies confirm the contributions of coverage, quality, and novelty, and demonstrate robustness to the choice of $K_{\max}$, significantly reducing the need for dataset-specific slot-count tuning in unsupervised object-centric learning.

Abstract

Slot Attention, an approach that binds different objects in a scene to a set of "slots", has become a leading method in unsupervised object-centric learning. Most methods assume a fixed slot count K, and to better accommodate the dynamic nature of object cardinality, a few works have explored K-adaptive variants. However, existing K-adaptive methods still suffer from two limitations. First, they do not explicitly constrain slot-binding quality, so low-quality slots lead to ambiguous feature attribution. Second, adding a slot-count penalty to the reconstruction objective creates conflicting optimization goals between reducing the number of active slots and maintaining reconstruction fidelity. As a result, they still lag significantly behind strong K-fixed baselines. To address these challenges, we propose Quality-Guided K-Adaptive Slot Attention (QASA). First, we decouple slot selection from reconstruction, eliminating the mutual constraints between the two objectives. Then, we propose an unsupervised Slot-Quality metric to assess per-slot quality, providing a principled signal for fine-grained slot--object binding. Based on this metric, we design a Quality-Guided Slot Selection scheme that dynamically selects a subset of high-quality slots and feeds them into our newly designed gated decoder for reconstruction during training. At inference, token-wise competition on slot attention yields a K-adaptive outcome. Experiments show that QASA substantially outperforms existing K-adaptive methods on both real and synthetic datasets. Moreover, on real-world datasets QASA surpasses K-fixed methods.

QASA: Quality-Guided K-Adaptive Slot Attention for Unsupervised Object-Centric Learning

TL;DR

This work addresses the challenge of varying object counts in unsupervised object-centric learning by introducing Quality-Guided K-Adaptive Slot Attention (QASA). It defines a Slot-Quality metric and a greedy, coverage-driven Quality-Guided Slot Selection to pick high-quality slots for reconstruction, while decoupling slot selection from reconstruction with a gated decoder. The approach supports both Transformer and MLP decoders and yields state-of-the-art results on real-world data (e.g., COCO, VOC) and competitive performance on synthetic data, outperforming existing -adaptive baselines by up to in mean best overlap and surpassing many -fixed methods. Ablation studies confirm the contributions of coverage, quality, and novelty, and demonstrate robustness to the choice of , significantly reducing the need for dataset-specific slot-count tuning in unsupervised object-centric learning.

Abstract

Slot Attention, an approach that binds different objects in a scene to a set of "slots", has become a leading method in unsupervised object-centric learning. Most methods assume a fixed slot count K, and to better accommodate the dynamic nature of object cardinality, a few works have explored K-adaptive variants. However, existing K-adaptive methods still suffer from two limitations. First, they do not explicitly constrain slot-binding quality, so low-quality slots lead to ambiguous feature attribution. Second, adding a slot-count penalty to the reconstruction objective creates conflicting optimization goals between reducing the number of active slots and maintaining reconstruction fidelity. As a result, they still lag significantly behind strong K-fixed baselines. To address these challenges, we propose Quality-Guided K-Adaptive Slot Attention (QASA). First, we decouple slot selection from reconstruction, eliminating the mutual constraints between the two objectives. Then, we propose an unsupervised Slot-Quality metric to assess per-slot quality, providing a principled signal for fine-grained slot--object binding. Based on this metric, we design a Quality-Guided Slot Selection scheme that dynamically selects a subset of high-quality slots and feeds them into our newly designed gated decoder for reconstruction during training. At inference, token-wise competition on slot attention yields a K-adaptive outcome. Experiments show that QASA substantially outperforms existing K-adaptive methods on both real and synthetic datasets. Moreover, on real-world datasets QASA surpasses K-fixed methods.
Paper Structure (13 sections, 11 equations, 4 figures, 6 tables, 1 algorithm)

This paper contains 13 sections, 11 equations, 4 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Method
  4. Overview
  5. Quality-Guided Slot Selection
  6. Gated Decoder
  7. Experiments
  8. Setup
  9. Main Results
  10. Comparison with $K$-adaptive methods.
  11. Comparison with $K$-fixed methods.
  12. Ablation Study
  13. Conclusion

Figures (4)

  • Figure 1: mBOi vs. slot count $K$ on COCO lin2014microsoft.$K$-fixed methods (e.g., SPOT kakogeorgiou2024spot, DINOSAUR seitzer2023bridging) exhibit large performance fluctuations as $K$ varies. By contrast, $K$-adaptive methods do not require searching for the best $K$. Our method QASA substantially outperforms existing $K$-adaptive baselines (MetaSlot liu2025metaslot, AdaSlot fan2024adaptive) and even surpasses strong $K$-fixed baselines.
  • Figure 2: Framework of Quality-Guided $K$-Adaptive Slot Attention (QASA).The overall training pipeline follows the canonical Slot Attention architecture: an image is encoded by the encoder, fed into the Slot Attention module, and then decoded; the reconstruction loss is computed against embeddings produced by a frozen encoder. The key difference is that we compute a slot-quality metric from the Slot Attention maps $A^{\text{slot}}$ and use it to select a subset of slots, allowing only the selected slots to participate in decoding. At inference, no slot selection is applied. Instead, all slots compete, based on $A^{\text{slot}}$, for each token to produce a $K$-adaptive result.
  • Figure 3: Visualizations on COCO. The AdaSlot examples are taken from its original paper fan2024adaptive. The MetaSlot examples are generated using the authors' released checkpoint liu2025metaslot. "SPOT-7" and "SPOT-20" denote SPOT kakogeorgiou2024spot with $K=7$ and $K=20$, respectively. The SPOT-7 examples are generated using the authors' released checkpoint. The SPOT-20 examples are generated by training a model under the same settings as in the original paper.
  • Figure 4: Visualizations on MOVi-C/E. The AdaSlot examples are taken from its original paper fan2024adaptive. the SPOT examples are generated by training a model under the same settings as in the original paper. MetaSlot has not released code for MOVi-C/E, so these results are not included in our comparison.