mlsanityData to 14 Oct 2026

Saccader: Improving Accuracy of Hard Attention Models for Vision

arXiv 1908.07644 ↗First posted 20 Aug 2019cs.CV cs.LG stat.MLcited by 35

Not on the map. Why?

Abstract (arXiv)

Although deep convolutional neural networks achieve state-of-the-art performance across nearly all image classification tasks, their decisions are difficult to interpret. One approach that offers some level of interpretability by design is textithard attention, which uses only relevant portions of the image. However, training hard attention models with only class label supervision is challenging, and hard attention has proved difficult to scale to complex datasets. Here, we propose a novel hard attention model, which we term Saccader. Key to Saccader is a pretraining step that requires only class labels and provides initial attention locations for policy gradient optimization. Our best models narrow the gap to common ImageNet baselines, achieving 75% top-1 and 91% top-5 while attending to less than one-third of the image.

Key facts. Cited by 35 papers in the map (data through 14 Oct 2026). First posted on arXiv 20 Aug 2019.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2021: 8; 2022: 6; 2023: 4; 2024: 7; 2025: 3; 2026: 7.

35 papers.

Sort:

By area:

  1. 2021Differentiable Patch Selection for Image Recognitioncited by 68
  2. 2021Vision Transformer with Progressive Samplingcited by 44
  3. 2021This looks more like that: Enhancing Self-Explaining Models by Prototypical Relevance Propagationcited by 43
  4. 2022Interpretable by Design: Learning Predictors by Composing Interpretable Queriescited by 28
  5. 2021Hard-Attention for Scalable Image Classificationcited by 18
  6. 2022AOE-Net: Entities Interactions Modeling with Adaptive Attention Mechanism for Temporal Action Proposals Generationcited by 14
  7. 2023Internally Rewarded Reinforcement Learningcited by 14
  8. 2021A Probabilistic Hard Attention Model For Sequentially Observed Scenescited by 12
  9. 2021AEI: Actors-Environment Interaction with Adaptive Attention for Temporal Action Proposals Generationcited by 11
  10. 2022Consistency driven Sequential Transformers Attention Model for Partially Observable Scenescited by 11
  11. 2023Foveation in the Era of Deep Learningcited by 8
  12. 2024AdaGlimpse: Active Visual Exploration with Arbitrary Glimpse Position and Scalecited by 7
23 more
  1. 2022A Lightweight, Efficient and Explainable-by-Design Convolutional Neural Network for Internet Traffic Classificationcited by 6
  2. 2024GQHAN: A Grover-inspired Quantum Hard Attention Networkcited by 5
  3. 2021Mask or Non-Mask? Robust Face Mask Detector via Triplet-Consistency Representation Learningcited by 3
  4. 2024Towards Two-Stream Foveation-based Active Vision Learningcited by 3
  5. 2025Emergence of Fixational and Saccadic Movements in a Multi-Level Recurrent Attention Model for Visioncited by 3
  6. 2021Recurrent Attention Models with Object-centric Capsule Representation for Multi-object Recognitioncited by 2
  7. 2026FOVI: A biologically-inspired foveated interface for deep vision modelscited by 2
  8. 2022GliTr: Glimpse Transformers with Spatiotemporal Consistency for Online Action Predictioncited by 1
  9. 2024Semantic-Syntactic Discrepancy in Images (SSDI): Learning Meaning and Order of Features from Natural Imagescited by 1
  10. 2024On Inherent Adversarial Robustness of Active Vision Systemscited by 1
  11. 2024Semantic-Based Active Perception for Humanoid Visual Tasks with Foveal Sensorscited by 1
  12. 2024TRIP: Trainable Region-of-Interest Prediction for Hardware-Efficient Neuromorphic Processing on Event-based Visioncited by 1
  13. 2026CanViT: Toward Active-Vision Foundation Modelscited by 1
  14. 2022Adaptive sampling for scanning pixel camerascited by 0
  15. 2023Energy-Efficient Visual Search by Eye Movement and Low-Latency Spiking Neural Networkcited by 0
  16. 2023Language-Guided Reinforcement Learning for Hard Attention in Few-Shot Learningcited by 0
  17. 2025A Sensorimotor Vision Transformercited by 0
  18. 2025SAFER-AiD: Saccade-Assisted Foveal-peripheral vision Enhanced Reconstruction for Adversarial Defensecited by 0
  19. 2026A Saccade-inspired Approach to Image Classification using Vision Transformer Attention Mapscited by 0
  20. 2026EVA: Bridging Performance and Human Alignment in Hard-Attention Vision Models for Image Classificationcited by 0
  21. 2026Saccade Attention Networks: Using Transfer Learning of Attention to Reduce Network Sizescited by 0
  22. 2026HiResNets: Native Full-HD Video Recognition with Foveal Residual Streamscited by 0
  23. 2026Task-driven Processing with Coarse-to-Fine Glimpse-based Active Perceptioncited by 0