Recent work has argued that classification losses utilizing softmax cross-entropy are superior not only for fixed-set classification tasks, but also by outperforming losses developed specifically for open-set tasks including few-shot learning and retrieval. Softmax classifiers have been studied using different embedding geometries—Euclidean, hyperbolic, and spherical—and claims have been made about the superiority of one or another, but they have not been systematically compared with careful controls. We conduct an empirical investigation of embedding geometry on soft-max losses for a variety of fixed-set classification and image retrieval tasks. An interesting property observed for the spherical losses lead us to propose a probabilistic classifier based on the von Mises–Fisher distribution, and we show that it is competitive with state-of-the-art methods while producing improved out-of-the-box calibration. We provide guidance regarding the trade-offs between losses and how to choose among them.

We present a pipeline for geomorphological analysis that uses structure from motion (SfM) and deep learning on close-range aerial imagery to estimate spatial distributions of rock traits (size, roundness, and orientation) along a tectonic fault scarp. The properties of the rocks on the fault scarp derive from the combination of initial volcanic fracturing and subsequent tectonic and geomorphic fracturing, and our pipeline allows scientists to leverage UAS-based imagery to gain a better understanding of such surface processes. We start by using SfM on aerial imagery to produce georeferenced orthomosaics and digital elevation models (DEM). A human expert then annotates rocks on a set of image tiles sampled from the orthomosaics, and these annotations are used to train a deep neural network to detect and segment individual rocks in the entire site. The extracted semantic information (rock masks) on large volumes of unlabeled, high-resolution SfM products allows subsequent structural analysis and shape descriptors to estimate rock size, roundness, and orientation. We present results of two experiments conducted along a fault scarp in the Volcanic Tablelands near Bishop, California. We conducted the first, proof-of-concept experiment with a DJI Phantom 4 Pro equipped with an RGB camera and inspected if elevation information assisted instance segmentation from RGB channels. Rock-trait histograms along and across the fault scarp were obtained with the neural network inference. In the second experiment, we deployed a hexrotor and a multispectral camera to produce a DEM and five spectral orthomosaics in red, green, blue, red edge, and near infrared. We focused on examining the effectiveness of different combinations of input channels in instance segmentation.

The focus in machine learning has branched beyond training classifiers on a single task to investigating how previously acquired knowledge in a source domain can be leveraged to facilitate learning in a related target domain, known as inductive transfer learning. Three active lines of research have independently explored transfer learning using neural networks. In weight transfer, a model trained on the source domain is used as an initialization point for a network to be trained on the target domain. In deep metric learning, the source domain is used to construct an embedding that captures class structure in both the source and target domains. In few-shot learning, the focus is on generalizing well in the target domain based on a limited number of labeled examples. We compare state-of-the-art methods from these three paradigms and also explore hybrid adapted-embedding methods that use limited target-domain data to fine tune embeddings constructed from source-domain data. We conduct a systematic comparison of methods in a variety of domains, varying the number of labeled instances available in the target domain (k), as well as the number of target-domain classes. We reach three principal conclusions: (1) Deep embeddings are far superior, compared to weight transfer, as a starting point for inter-domain transfer or model re-use (2) Our hybrid methods robustly outperform every few-shot learning and every deep metric learning method previously proposed, with a mean error reduction of 34% over state-of-the-art. (3) Among loss functions for discovering embeddings, the histogram loss (Ustinova & Lempitsky, 2016) is most robust. We hope our results will motivate a unification of research in weight transfer, deep metric learning, and few-shot learning.