This study introduces a novel expert generation method that dynamically reduces task and computational complexity without compromising predictive performance. It is based on a new hierarchical classification network topology that combines sequential processing of generic low-level features with parallelism and nesting of high-level features. This structure allows for the innovative extraction technique: the ability to select only high-level features of task-relevant categories. In certain cases, it is possible to skip almost all unneeded high-level features, which can significantly reduce the inference cost and is highly beneficial in resource-constrained conditions. We believe this method paves the way for future network designs that are lightweight and adaptable, making them suitable for a wide range of applications, from compact edge devices to large-scale clouds. In terms of dynamic inference our methodology can achieve an exclusion of up to 88.7\,\% of parameters and 73.4\,\% fewer giga-multiply accumulate (GMAC) operations, analysis against comparative baselines showing an average reduction of 47.6\,\% in parameters and 5.8\,\% in GMACs across the cases we evaluated.
Crop detection is integral for precision agriculture applications such as automated yield estimation or fruit picking. However, crop detection, e.g., apple detection in orchard environments remains challenging due to a lack of large-scale datasets and the small relative size of the crops in the image. In this work, we address these challenges by reformulating the apple detection task in a semi-supervised manner. To this end, we provide the large, high-resolution dataset MAD comprising 105 labeled images with 14,667 annotated apple instances and 4,440 unlabeled images. Utilizing this dataset, we also propose a novel Semi-Supervised Small Apple Detection system S$^3$AD based on contextual attention and selective tiling to improve the challenging detection of small apples, while limiting the computational overhead. We conduct an extensive evaluation on MAD and the MSU dataset, showing that S$^3$AD substantially outperforms strong fully-supervised baselines, including several small object detection systems, by up to $14.9\%$. Additionally, we exploit the detailed annotations of our dataset w.r.t. apple properties to analyze the influence of relative size or level of occlusion on the results of various systems, quantifying current challenges.
In this work, we parallelize high-level features in deep networks to selectively skip or select class-specific features to reduce inference costs. This challenges most deep learning methods due to their limited ability to efficiently and effectively focus on selected class-specific features without retraining. We propose a serial-parallel hybrid architecture with serial generic low-level features and parallel high-level features. This accounts for the fact that many high-level features are class-specific rather than generic, and has connections to recent neuroscientific findings that observe spatially and contextually separated neural activations in the human brain. Our approach provides the unique functionality of cutouts: selecting parts of the network to focus on only relevant subsets of classes without requiring retraining. High performance is maintained, but the cost of inference can be significantly reduced. In some of our examples, up to $75\,\%$ of parameters are skipped and $35\,\%$ fewer GMACs (Giga multiply-accumulate) operations are used as the approach adapts to a change in task complexity. This is important for mobile, industrial, and robotic applications where reducing the number of parameters, the computational complexity, and thus the power consumption can be paramount. Another unique functionality is that it allows processing to be directly influenced by enhancing or inhibiting high-level class-specific features, similar to the mechanism of selective attention in the human brain. This can be relevant for cross-modal applications, the use of semantic prior knowledge, and/or context-aware processing.
Traffic light detection is a challenging problem in the context of self-driving cars and driver assistance systems. While most existing systems produce good results on large traffic lights, detecting small and tiny ones is often overlooked. A key problem here is the inherent downsampling in CNNs, leading to low-resolution features for detection. To mitigate this problem, we propose a new traffic light detection system, comprising a novel traffic light proposal generator that utilizes findings from general object proposal generation, fine-grained multi-scale features, and attention for efficient processing. Moreover, we design a new detection head for classifying and refining our proposals. We evaluate our system on three challenging, publicly available datasets and compare it against six methods. The results show substantial improvements of at least $12.6\%$ on small and tiny traffic lights, as well as strong results across all sizes of traffic lights.
This paper introduces an audio-visual speech enhancement system that leverages score-based generative models, also known as diffusion models, conditioned on visual information. In particular, we exploit audio-visual embeddings obtained from a self-super\-vised learning model that has been fine-tuned on lipreading. The layer-wise features of its transformer-based encoder are aggregated, time-aligned, and incorporated into the noise conditional score network. Experimental evaluations show that the proposed audio-visual speech enhancement system yields improved speech quality and reduces generative artifacts such as phonetic confusions with respect to the audio-only equivalent. The latter is supported by the word error rate of a downstream automatic speech recognition model, which decreases noticeably, especially at low input signal-to-noise ratios.
We investigate cross-quality knowledge distillation (CQKD), a knowledge distillation method where knowledge from a teacher network trained with full-resolution images is transferred to a student network that takes as input low-resolution images. As image size is a deciding factor for the computational load of computer vision applications, CQKD notably reduces the requirements by only using the student network at inference time. Our experimental results show that CQKD outperforms supervised learning in large-scale image classification problems. We also highlight the importance of calibrating neural networks: we show that with higher temperature smoothing of the teacher's output distribution, the student distribution exhibits a higher entropy, which leads to both, a lower calibration error and a higher network accuracy.
Neural radiance field (NeRF), in particular its extension by instant neural graphics primitives, is a novel rendering method for view synthesis that uses real-world images to build photo-realistic immersive virtual scenes. Despite its potential, research on the combination of NeRF and virtual reality (VR) remains sparse. Currently, there is no integration into typical VR systems available, and the performance and suitability of NeRF implementations for VR have not been evaluated, for instance, for different scene complexities or screen resolutions. In this paper, we present and evaluate a NeRF-based framework that is capable of rendering scenes in immersive VR allowing users to freely move their heads to explore complex real-world scenes. We evaluate our framework by benchmarking three different NeRF scenes concerning their rendering performance at different scene complexities and resolutions. Utilizing super-resolution, our approach can yield a frame rate of 30 frames per second with a resolution of 1280x720 pixels per eye. We discuss potential applications of our framework and provide an open source implementation online.
Precisely locating and segmenting medical instruments in images of minimally invasive surgeries, medical instrument segmentation, is an essential first step for several tasks in medical image processing. However, image degradations, small instruments, and the generalization between different surgery types make medical instrument segmentation challenging. To cope with these challenges, we adapt the object proposal generation system AttentionMask and propose a dedicated post-processing to select promising proposals. The results on the ROBUST-MIS Challenge 2019 show that our adapted AttentionMask system is a strong foundation for generating state-of-the-art performance. Our evaluation in an object proposal generation framework shows that our adapted AttentionMask system is robust to image degradations, generalizes well to unseen types of surgeries, and copes well with small instruments.
The localization of fruits is an essential first step in automated agricultural pipelines for yield estimation or fruit picking. One example of this is the localization of apples in images of entire apple trees. Since the apples are very small objects in such scenarios, we tackle this problem by adapting the object proposal generation system AttentionMask that focuses on small objects. We adapt AttentionMask by either adding a new module for very small apples or integrating it into a tiling framework. Both approaches clearly outperform standard object proposal generation systems on the MinneApple dataset covering complex apple orchard environments. Our evaluation further analyses the improvement w.r.t. the apple sizes and shows the different characteristics of our two approaches.
Class-agnostic object proposal generation is an important first step in many object detection pipelines. However, object proposals of modern systems are rather inaccurate in terms of segmentation and only roughly adhere to object boundaries. Since typical refinement steps are usually not applicable to thousands of proposals, we propose a superpixel-based refinement system for object proposal generation systems. Utilizing precise superpixels and superpixel pooling on deep features, we refine initial coarse proposals in an end-to-end learned system. Furthermore, we propose a novel DeepFH segmentation, which enriches the classic Felzenszwalb and Huttenlocher (FH) segmentation with deep features leading to improved segmentation results and better object proposal refinements. On the COCO dataset with LVIS annotations, we show that our refinement based on DeepFH superpixels outperforms state-of-the-art methods and leads to more precise object proposals.