Long-tailed Species Recognition in the NACTI Wildlife Dataset

2025-10-24

AI Summary

The North American Camera Trap Image (NACTI) dataset exhibits severe long-tailed class imbalance, with head classes dominating and tail classes—particularly rare wildlife species—represented by very few samples. Method: We systematically evaluate and enhance long-tailed recognition paradigms, proposing an optimized training framework integrating weighted cross-entropy (WCE) loss, sensitivity regularization, and adaptive learning rate scheduling, implemented end-to-end on the PyTorch Wildlife platform. Results: Our method achieves 99.40% Top-1 accuracy on the standard NACTI test set (+3.89 points over baseline) and, more critically, attains 52.55% accuracy on our newly constructed de-biased test set (ENA-Detection), demonstrating substantially improved cross-domain generalization. Experiments expose fundamental limitations of existing approaches under extreme domain shift. This work establishes a reproducible, robust new baseline for fine-grained, long-tailed wildlife recognition.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Algorithmic accountability and transparency on the web
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

As most''in the wild''data collections of the natural world, the North America Camera Trap Images (NACTI) dataset shows severe long-tailed class imbalance, noting that the largest'Head'class alone covers>50% of the 3.7M images in the corpus. Building on the PyTorch Wildlife model, we present a systematic study of Long-Tail Recognition methodologies for species recognition on the NACTI dataset covering experiments on various LTR loss functions plus LTR-sensitive regularisation. Our best configuration achieves 99.40% Top-1 accuracy on our NACTI test data split, substantially improving over a 95.51% baseline using standard cross-entropy with Adam. This also improves on previously reported top performance in MLWIC2 at 96.8% albeit using partly unpublished (potentially different) partitioning, optimiser, and evaluation protocols. To evaluate domain shifts (e.g. night-time captures, occlusion, motion-blur) towards other datasets we construct a Reduced-Bias Test set from the ENA-Detection dataset where our experimentally optimised long-tail enhanced model achieves leading 52.55% accuracy (up from 51.20% with WCE loss), demonstrating stronger generalisation capabilities under distribution shift. We document the consistent improvements of LTR-enhancing scheduler choices in this NACTI wildlife domain, particularly when in tandem with state-of-the-art LTR losses. We finally discuss qualitative and quantitative shortcomings that LTR methods cannot sufficiently address, including catastrophic breakdown for'Tail'classes under severe domain shift. For maximum reproducibility we publish all dataset splits, key code, and full network weights.

Problem

• Addressing severe class imbalance in wildlife species recognition datasets
• Improving recognition accuracy for long-tailed species distributions
• Enhancing model generalization under domain shifts and biases

Innovation

• Long-tail recognition loss functions for species classification
• LTR-sensitive regularization to handle class imbalance
• Domain shift adaptation with reduced-bias test evaluation