Classification performance based on ImageNet is the de-facto standard metric for CNN development. In this work we challenge the notion that CNN architecture design solely based on ImageNet leads to generally effective convolutional neural network (CNN) architectures that perform well on a diverse set of datasets and application domains. To this end, we investigate and ultimately improve ImageNet as a basis for deriving such architectures. We conduct an extensive empirical study for which we train 500 CNN architectures, sampled from the broad AnyNetX design space, on ImageNet as well as 8 additional well-known image classification benchmark datasets from a diverse array of application domains. We observe that the performances of the architectures are highlydataset dependent. Some datasets even exhibit a negative error correlation with ImageNet across all architectures. We show how to significantly increase these correlations byutilizing ImageNet subsets restricted to fewer classes. These contributions can have a profound impact on the way we design future CNN architectures and help alleviate the tilt we see currently in our community with respect to over-reliance on one dataset.

We present solutions to two of the most pressing issues in contemporary optical music recognition (OMR). We improve recognition accuracy on low-quality, real-world (i.e. containing ageing, lighting, or dirt artefacts among others) input data and provide confidence-rated model outputs to enable efficient human post-processing. Specifically, we present (i) a sophisticated input augmentation scheme that can reduce the gap between sanitised benchmarks and realistic tasks through a combination of synthetic data and noisy perturbations of real-world documents; (ii) an adversarial discriminative domain adaptation method that can be employed to improve the performance of OMR systems on low-quality data; (iii) a combination of model ensembles and prediction fusion, which generates trustworthy confidence ratings for each prediction. We evaluate our contributions on a newly created test set consisting of manually annotated pages of varying real-world quality, sourced from the International Music Score Library Project (IMSLP)/Petrucci Music Library. With the presented data augmentation scheme, we achieve a doubling in detection performance from 36.0% to 73.3% on noisy real-world data compared to state-of-the-art training. This result is then combined with robust confidence ratings paving the way for OMR to be deployed in the real world. Additionally, we show the merits of unsupervised adversarial domain adaptation for OMR raising the 36.0% baseline to 48.9%. All our code and data are freely available at: https://github.com/raember/s2anet/tree/TISMIR_publication.

We present an extensive evaluation of a wide variety of promising design patterns for automated deep-learning (AutoDL) methods, organized according to the problem categories of the 2019 AutoDL challenges, which set the task of optimizing both model accuracy and search efficiency under tight time and computing constraints. We propose structured empirical evaluations as the most promising avenue to obtain design principles for deep-learning systems due to the absence of strong theoretical support. From these evaluations, we distill relevant patterns which give rise to neural network design recommendations. In particular, we establish (a) that very wide fully connected layers learn meaningful features faster; we illustrate (b) how the lack of pretraining in audio processing can be compensated by architecture search; we show (c) that in text processing deep-learning-based methods only pull ahead of traditional methods for short text lengths with less than a thousand characters under tight resource limitations; and lastly we present (d) evidence that in very data- and computing-constrained settings, hyperparameter tuning of more traditional machine-learning methods outperforms deep-learning systems.

Humans and animals recognize objects irrespective of the beholder's point of view, which may drastically change their appearances. Artificial pattern recognizers also strive to achieve this, e.g., through translational invariance in convolutional neural networks (CNNs). However, both CNNs and vision transformers (ViTs) perform very poorly on rotated inputs. Here we present artificial mental rotation (AMR), a novel deep learning paradigm for dealing with in-plane rotations inspired by the neuro-psychological concept of mental rotation. Our simple AMR implementation works with all common CNN and ViT architectures. We test it on ImageNet, Stanford Cars, and Oxford Pet. With a top-1 error (averaged across datasets and architectures) of $0.743$, AMR outperforms the current state of the art (rotational data augmentation, average top-1 error of $0.626$) by $19\%$. We also easily transfer a trained AMR module to a downstream task to improve the performance of a pre-trained semantic segmentation model on rotated CoCo from $32.7$ to $55.2$ IoU.

This paper gives an overview of our current Optical Music Recognition (OMR) research. We recently released the OMR dataset \emph{DeepScores} as well as the object detection method \emph{Deep Watershed Detector}. We are currently taking some additional steps to improve both of them. Here we summarize current and future efforts, aimed at improving usefulness on real-world task and tackling extreme class imbalance.