FlexiGrad: Adaptive Gradient Modulation for Hierarchical Fine-Grained Classification

2026-07-20

AI Summary

This work addresses the training instability and representation inconsistency in hierarchical fine-grained classification caused by gradient conflicts between coarse- and fine-grained tasks. To resolve this, the authors propose FlexiGrad, a method that dynamically preserves shared gradient directions across multiple levels while suppressing conflicting components during backpropagation through gradient decomposition and adaptive weighting. FlexiGrad requires no additional parameters or architectural modifications and incorporates a smooth, hierarchy-aware weighting function to harmonize multi-granularity supervision signals, making it compatible with mainstream deep architectures. Experimental results demonstrate that FlexiGrad significantly improves multi-granularity classification performance on benchmark datasets including CUB-200-2011, FGVC-Aircraft, and Stanford Cars.

Abstract

Many fine-grained recognition tasks contain hierarchical labels such as order, family and species. Although this supervision should be beneficial, jointly optimising all levels often leads to unstable training because coarse and fine classifiers impose inconsistent gradients on the shared backbone. This hierarchical gradient conflict prevents the model from learning a coherent coarse-to-fine representation. In this paper, we propose FlexiGrad, a simple and parameter-free method that regulates gradient interactions during backpropagation. FlexiGrad removes only the harmful conflicting component when tasks disagree and reinforces the shared direction when they partially agree through a smooth hierarchy-aware weighting function. This produces stable optimisation and preserves both global structure and fine-grained discriminative cues. FlexiGrad integrates into existing architectures without modification while improves multi-granularity accuracy on CUB-200-2011, FGVC-Aircraft and Stanford Cars. The code will be available at PRIS-CV/FlexiGrad.

Problem

• hierarchical classification
• fine-grained recognition
• gradient conflict
• multi-granularity learning
• coarse-to-fine representation

Innovation

• gradient modulation
• hierarchical classification
• fine-grained recognition
• multi-task learning
• parameter-free