Presence-Only Geographical Priors for Fine-Grained Image Classification

Abstract

Appearance information alone is often not sufficient to accurately differentiate between fine-grained visual categories. Human experts make use of additional cues such as where, and when, a given image was taken in order to inform their final decision. This contextual information is readily available in many online image collections but has been underutilized by existing image classifiers that focus solely on making predictions based on the image contents.We propose an efficient spatio-temporal prior, that when conditioned on a geographical location and time, estimates the probability that a given object category occurs at that location. Our prior is trained from presence-only observation data and jointly models object categories, their spatiotemporal distributions, and photographer biases. Experiments performed on multiple challenging image classification datasets show that combining our prior with the predictions from image classifiers results in a large improvement in final classification performance.

Thinking like a naturalist: Enhancing computer vision of citizen science images by harnessing contextual data

“…iRecord (http://www.brc.ac.uk/irecord) partially automates this process, flagging records to expert verifiers that are labelled as being outside of the known range. Distribution priors have been shown to be effective in improving the identification of North American birds (Berg et al, ), images in the iNaturalist dataset (Mac Aodha, Cole, & Perona, ) and generating location‐specific shortlists of German plants (Wittich, Seeland, Wäldchen, Rzanny, & Mäder, ). This approach can greatly reduce the risk of non‐sensical identifications that otherwise lead to considerable scepticism over the use of automated methods (Gaston & O'Neill, ).…”

Toward a Yearly Country-Scale CORINE Land-Cover Map without Using Images: A Map Translation Approach

“…Thus, we decide to incorporate into our network a positional encoding sub-module to take into account the coordinates of each patch during the translation process. Previous attempts have been made to improve CNN-based classification of terrestrial images using various embedded features [59]. The main strategy includes a direct insertion of (quantized) latitudes and longitudes [60] or to retrieve proxy attributes that are more likely to be discriminative (population statistics, social media, see [61,62].…”

Scaling Biodiversity Monitoring for the Data Age

“…We can aggregate information from remote sensing, passive and active monitoring sensors, ecological samples, and the natural history record to paint a cohesive picture of global biodiversity and help fight to protect it. species identification by letting the model know which species are most likely to be seen in a given area at a specific time [8], and building models that can share information across data collected by a given static sensor, helping the model adapt to previously unseen environments [9]. Aggregating data allows researchers to share the cost and scale up, in collection effort, data processing effort, and across jurisdictions.…”