As with most ``in the wild'' collections of the natural world, the North America Camera Trap Images (NACTI) dataset exhibits long-tailed class imbalance, with the largest class covering over 50% of its 3.7M images. Building on the PyTorch Wildlife model, we systematically evaluate Long-Tail Recognition (LTR) methodologies to benchmark species recognition performance, including specialised loss functions and LTR-sensitive regularisation. Our optimised configuration achieves state-of-the-art 99.40% Top-1 accuracy on the NACTI test split, significantly outperforming standard baselines and previously reported top performances. To assess robustness under domain shifts (e.g., night-time captures, occlusion, motion-blur), we extend our evaluation across three independent reduced-bias test sets (including ENA-Detection, Caltech Camera Traps and Missouri Camera Traps). Across these out-of-distribution (OOD) evaluations, our LTR-enhanced model consistently demonstrates substantially stronger generalisation capabilities compared to standard cross-entropy approaches. However, qualitative and quantitative analyses underline that current LTR optimisations cannot fully overcome representational bottlenecks, resulting in catastrophic predictive breakdown for rare `Tail' classes under severe domain shift. For maximum reproducibility, all dataset splits, key code, and network weights are published with this paper at https://github.com/ZehuaLiuY/Species-Classification.
Camera traps often run in the field on edge hardware with limited or no connectivity, making small, locally-deployable vision-language models (VLMs) -- not frontier-scale ones -- the practically relevant class to evaluate for species identification. We test whether models in this deployment-relevant 2--8B range carry genuine taxonomic knowledge, evaluating four such VLMs (Qwen3-VL 2B/4B/8B, Gemma3 4B) against the domain-specific specialist BioCLIP (300M parameters) on a 96-species task, comparing clean iNaturalist photographs against camera-trap imagery from 6 LILA.science collections, on two independently-sampled evaluation sets. All models identify species far above chance, but every model -- general-purpose or specialist -- degrades sharply on field imagery (domain gaps of 9.6--26.6 percentage points, consistent across taxonomic levels and both evaluation sets), indicating the degradation reflects general image legibility rather than fine-grained discrimination failure. BioCLIP substantially outperforms every VLM tested (by 33.2--59.2 percentage points across an expanded 200-image sample for every model) despite its far smaller size, suggesting the gap reflects specialized training data rather than model scale; yet BioCLIP's own domain gap (18.0 points) is statistically indistinguishable from the best VLM's (22.3 points), suggesting the clean-to-field degradation itself is a property of the image-quality shift rather than a general-purpose-model weakness. Under open-set prompting, 5.9--9.6% of responses are syntactically valid but taxonomically nonexistent species names; the relative fabrication-rate ranking across models replicates exactly across both evaluation sets, a more robust finding than any single point estimate.
Long-tailed recognition suffers from a persistent head--tail trade-off: improving tail performance often degrades head accuracy and can increase training instability. Despite strong empirical results from re-weighting, decoupled training, and multi-expert methods, key design choices about representation sharing between head and tail classes and supervision weighting across class groups remain largely heuristic. In this work, we propose OSDTW, a principled task-decomposition framework that partitions the original single-label recognition problem into a head task and a tail task, implemented with a shared encoder and task-specific decoders. To handle the mutual exclusivity and statistical dependence between the two label groups, we introduce a factorized model and show that the resulting Kullback--Leibler divergence-based generalization error can be written as the sum of task-wise terms up to an additive constant, yielding a well-defined task-wise objective. We further develop a three-stage training pipeline: independent task training to estimate task-wise optima and the Fisher information matrix, weighted joint training to learn a shared encoder, and branch assembly to construct the final decoupled model. Under a block-diagonal Fisher approximation, we derive a computable second-order expansion of the expected generalization error, decomposing it into encoder variance, encoder bias, and decoder variance. This bias--variance decomposition provides a computable proxy to select the shared depth and task weights, enabling efficient hyper-parameter search. Experiments on standard long-tailed benchmarks demonstrate the effectiveness of the proposed approach over strong baselines.
Supervised deep learning methods enable the rapid processing of ecological image data, but depend on a costly annotation process. Consequently, training labels are commonly derived from volunteer citizen science projects. However, disagreement among volunteers introduces uncertainty in the "ground truth" data that are assumed to be correct for model training and validation. Using two datasets containing camera trap images with associated volunteer and expert classifications, we investigated the effects of training under higher ground truth uncertainty. We observed improved overall test accuracy, particularly for images that were more difficult for volunteers. Species-level accuracy also generally improved, but generalisation to a different dataset did not. The benefits of ground truth uncertainty were enhanced by pre-training on ImageNet. Pre-training also reduced the number of training epochs required; further reductions in computational cost, but not gains in accuracy, resulted from additional pre-training on other camera trap images. With unbalanced training data, we still observed a clear benefit of increased ground truth uncertainty for overall accuracy, especially on difficult images. Class imbalance improved accuracy for common species, reduced rare species accuracy, and changed patterns of misclassification to more closely resemble mistakes made by volunteers. Our findings have implications for applying deep learning across ecological image types with multiple labels. Practitioners can improve accuracy, especially on difficult examples, by including moderate levels of label disagreement during training and using models pre-trained on general image data. In addition to improving the use of citizen science-derived labels in model training, our study suggests avenues for more effectively integrating human and deep learning classifications in combined workflows. (abridged)
Leonard Hockerts, Peter S. Stewart, Sarthak Arora +1