TasksSotADatasetsPapersMethodsSubmitAbout
Papers With Code 2

A community resource for machine learning research: papers, code, benchmarks, and state-of-the-art results.

Explore

Notable BenchmarksAll SotADatasetsPapersMethods

Community

Submit ResultsAbout

Data sourced from the PWC Archive (CC-BY-SA 4.0). Built by the community, for the community.

Papers/Adaptive Parametric Activation

Adaptive Parametric Activation

Konstantinos Panagiotis Alexandridis, Jiankang Deng, Anh Nguyen, Shan Luo

2024-07-11Long-tail LearningInstance Segmentationimbalanced classification
PaperPDFCode(official)

Abstract

The activation function plays a crucial role in model optimisation, yet the optimal choice remains unclear. For example, the Sigmoid activation is the de-facto activation in balanced classification tasks, however, in imbalanced classification, it proves inappropriate due to bias towards frequent classes. In this work, we delve deeper in this phenomenon by performing a comprehensive statistical analysis in the classification and intermediate layers of both balanced and imbalanced networks and we empirically show that aligning the activation function with the data distribution, enhances the performance in both balanced and imbalanced tasks. To this end, we propose the Adaptive Parametric Activation (APA) function, a novel and versatile activation function that unifies most common activation functions under a single formula. APA can be applied in both intermediate layers and attention layers, significantly outperforming the state-of-the-art on several imbalanced benchmarks such as ImageNet-LT, iNaturalist2018, Places-LT, CIFAR100-LT and LVIS and balanced benchmarks such as ImageNet1K, COCO and V3DET. The code is available at https://github.com/kostas1515/AGLU.

Results

TaskDatasetMetricValueModel
Image ClassificationPlaces-LTTop-1 Accuracy42APA (SE-ResNet-50)
Image ClassificationiNaturalist 2018Top-1 Accuracy74.8APA (SE-ResNet-50)
Image ClassificationImageNet-LTTop-1 Accuracy59.1APA (SE-ResNext-50)
Image ClassificationImageNet-LTTop-1 Accuracy57.9APA (SE-ResNet-50)
Instance SegmentationLVIS v1.0 valmask AP30.7SE-R101-FPN-MaskRCNN-APA
Instance SegmentationLVIS v1.0 valmask APr23.6SE-R101-FPN-MaskRCNN-APA
Instance SegmentationLVIS v1.0 valmask AP29.1SE-R50-FPN-MaskRCNN-APA
Instance SegmentationLVIS v1.0 valmask APr21.6SE-R50-FPN-MaskRCNN-APA
Few-Shot Image ClassificationPlaces-LTTop-1 Accuracy42APA (SE-ResNet-50)
Few-Shot Image ClassificationiNaturalist 2018Top-1 Accuracy74.8APA (SE-ResNet-50)
Few-Shot Image ClassificationImageNet-LTTop-1 Accuracy59.1APA (SE-ResNext-50)
Few-Shot Image ClassificationImageNet-LTTop-1 Accuracy57.9APA (SE-ResNet-50)
Generalized Few-Shot ClassificationPlaces-LTTop-1 Accuracy42APA (SE-ResNet-50)
Generalized Few-Shot ClassificationiNaturalist 2018Top-1 Accuracy74.8APA (SE-ResNet-50)
Generalized Few-Shot ClassificationImageNet-LTTop-1 Accuracy59.1APA (SE-ResNext-50)
Generalized Few-Shot ClassificationImageNet-LTTop-1 Accuracy57.9APA (SE-ResNet-50)
Long-tail LearningPlaces-LTTop-1 Accuracy42APA (SE-ResNet-50)
Long-tail LearningiNaturalist 2018Top-1 Accuracy74.8APA (SE-ResNet-50)
Long-tail LearningImageNet-LTTop-1 Accuracy59.1APA (SE-ResNext-50)
Long-tail LearningImageNet-LTTop-1 Accuracy57.9APA (SE-ResNet-50)
Generalized Few-Shot LearningPlaces-LTTop-1 Accuracy42APA (SE-ResNet-50)
Generalized Few-Shot LearningiNaturalist 2018Top-1 Accuracy74.8APA (SE-ResNet-50)
Generalized Few-Shot LearningImageNet-LTTop-1 Accuracy59.1APA (SE-ResNext-50)
Generalized Few-Shot LearningImageNet-LTTop-1 Accuracy57.9APA (SE-ResNet-50)

Related Papers

SCORE: Scene Context Matters in Open-Vocabulary Remote Sensing Instance Segmentation2025-07-17Tomato Multi-Angle Multi-Pose Dataset for Fine-Grained Phenotyping2025-07-15Beyond Appearance: Geometric Cues for Robust Video Instance Segmentation2025-07-08SPADE: Spatial-Aware Denoising Network for Open-vocabulary Panoptic Scene Graph Generation with Long- and Local-range Context Reasoning2025-07-08DreamGrasp: Zero-Shot 3D Multi-Object Reconstruction from Partial-View Images for Robotic Manipulation2025-07-08No time to train! Training-Free Reference-Based Instance Segmentation2025-07-03NOCTIS: Novel Object Cyclic Threshold based Instance Segmentation2025-07-02VoteSplat: Hough Voting Gaussian Splatting for 3D Scene Understanding2025-06-28