Comparing fine-grained and coarse-grained object detection for ecology

TL;DR

This paper tackles how the granularity of species recognition in camera-trap data affects ecological interpretation. It compares fine-grained species-level and coarse-grained functional-group classifications using the Wild Deserts dataset and evaluates the impact of including negative samples. The approach leverages YOLOv8-based detectors and reports $mAP_{50}$ at IoU = 0.5 for both granularity levels, finding that coarse-grained performance is often superior for morphologically similar species, with mixed results for dissimilar species. The findings provide practical guidance for ecologists: group morphologically similar species when appropriate and cautiously apply negative samples, as gains are modest and context-dependent.

Abstract

Computer vision applications are increasingly popular for wildlife monitoring tasks. While some studies focus on the monitoring of a single species, such as a particular endangered species, others monitor larger functional groups, such as predators. In our study, we used camera trap images collected in north-western New South Wales, Australia, to investigate how model results were affected by combining multiple species in single classes, and whether the addition of negative samples can improve model performance. We found that species that benefited the most from merging into a single class were mainly species that look alike morphologically, i.e. macropods. Whereas species that looked distinctively different gave mixed results when merged, e.g. merging pigs and goats together as non-native large mammals. We also found that adding negative samples improved model performance marginally in most instances, and recommend conducting a more comprehensive study to explore whether the marginal gains were random or consistent. We suggest that practitioners could classify morphologically similar species together as a functional group or higher taxonomic group to draw ecological inferences. Nevertheless, whether to merge classes or not will depend on the ecological question to be explored.

Figures (4)

• Figure 1: The distribution of the Wild Deserts dataset and the number of instances of each species.
• Figure 2: Mean average precision (mAP) of models (a) without and (b) with negative samples. Columns represent (i) mAP of the fine-grained classes from the fine-grained model calculated by YOLOv8, (ii) coarse-grained mAP extracted from the fine-grained model with pycocotools, and (iii) coarse-grained mAP from the coarse-grained model calculated by YOLOv8.
• Figure 3: Examples of macropod species from the Wild Deserts dataset. From the left: Red kangaroo (Osphranter rufus), Western grey kangaroo (Macropus fuliginosus), and Euro (Osphranter robustus)
• Figure 4: Normalised confusion matrix of the fine-grained model before adding negative samples.