Abstract:The equivariant behaviour of features is essential in many computer vision tasks, yet popular self-supervised learning (SSL) methods tend to constrain equivariance by design. We propose a self-supervised learning approach where the system learns transformations independently by reconstructing images that have undergone previously unseen transformations. Specifically, the model is tasked to reconstruct intermediate transformed images, e.g. translated or rotated images, without prior knowledge of these transformations. This auxiliary task encourages the model to develop equivariance-coherent features without relying on predefined transformation rules. To this end, we apply transformations to the input image, generating an image pair, and then split the extracted features into two sets per image. One set is used with a usual SSL loss encouraging invariance, the other with our loss based on the auxiliary task to reconstruct the intermediate transformed images. Our loss and the SSL loss are linearly combined with weighted terms. Evaluating on synthetic tasks with natural images, our proposed method strongly outperforms all competitors, regardless of whether they are designed to learn equivariance. Furthermore, when trained alongside augmentation-based methods as the invariance tasks, such as iBOT or DINOv2, we successfully learn a balanced combination of invariant and equivariant features. Our approach performs strong on a rich set of realistic computer vision downstream tasks, almost always improving over all baselines.
Abstract:Cell tracking is a key computational task in live-cell microscopy, but fully automated analysis of high-throughput imaging requires reliable and, thus, uncertainty-aware data analysis tools, as the amount of data recorded within a single experiment exceeds what humans are able to overlook. We here propose and benchmark various methods to reason about and quantify uncertainty in linear assignment-based cell tracking algorithms. Our methods take inspiration from statistics and machine learning, leveraging two perspectives on the cell tracking problem explored throughout this work: Considering it as a Bayesian inference problem and as a classification problem. Our methods admit a framework-like character in that they equip any frame-to-frame tracking method with uncertainty quantification. We demonstrate this by applying it to various existing tracking algorithms including the recently presented Transformer-based trackers. We demonstrate empirically that our methods yield useful and well-calibrated tracking uncertainties.
Abstract:Parameter-Efficient Fine-Tuning (PEFT) methods have transformed the approach to fine-tuning large models for downstream tasks by enabling the adjustment of significantly fewer parameters than those in the original model matrices. In this work, we study the "very low rank regime", where we fine-tune the lowest amount of parameters per linear layer for each considered PEFT method. We propose 1LoRA (Summation Low-Rank Adaptation), a compute, parameter and memory efficient fine-tuning method which uses the feature sum as fixed compression and a single trainable vector as decompression. Differently from state-of-the-art PEFT methods like LoRA, VeRA, and the recent MoRA, 1LoRA uses fewer parameters per layer, reducing the memory footprint and the computational cost. We extensively evaluate our method against state-of-the-art PEFT methods on multiple fine-tuning tasks, and show that our method not only outperforms them, but is also more parameter, memory and computationally efficient. Moreover, thanks to its memory efficiency, 1LoRA allows to fine-tune more evenly across layers, instead of focusing on specific ones (e.g. attention layers), improving performance further.
Abstract:Magnetic Resonance Imaging (MRI) is a powerful imaging technique widely used for visualizing structures within the human body and in other fields such as plant sciences. However, there is a demand to develop fast 3D-MRI reconstruction algorithms to show the fine structure of objects from under-sampled acquisition data, i.e., k-space data. This emphasizes the need for efficient solutions that can handle limited input while maintaining high-quality imaging. In contrast to previous methods only using 2D, we propose a 3D MRI reconstruction method that leverages a regularized 3D diffusion model combined with optimization method. By incorporating diffusion based priors, our method improves image quality, reduces noise, and enhances the overall fidelity of 3D MRI reconstructions. We conduct comprehensive experiments analysis on clinical and plant science MRI datasets. To evaluate the algorithm effectiveness for under-sampled k-space data, we also demonstrate its reconstruction performance with several undersampling patterns, as well as with in- and out-of-distribution pre-trained data. In experiments, we show that our method improves upon tested competitors.
Abstract:Augmentation-based self-supervised learning methods have shown remarkable success in self-supervised visual representation learning, excelling in learning invariant features but often neglecting equivariant ones. This limitation reduces the generalizability of foundation models, particularly for downstream tasks requiring equivariance. We propose integrating an image reconstruction task as an auxiliary component in augmentation-based self-supervised learning algorithms to facilitate equivariant feature learning without additional parameters. Our method implements a cross-attention mechanism to blend features learned from two augmented views, subsequently reconstructing one of them. This approach is adaptable to various datasets and augmented-pair based learning methods. We evaluate its effectiveness on learning equivariant features through multiple linear regression tasks and downstream applications on both artificial (3DIEBench) and natural (ImageNet) datasets. Results consistently demonstrate significant improvements over standard augmentation-based self-supervised learning methods and state-of-the-art approaches, particularly excelling in scenarios involving combined augmentations. Our method enhances the learning of both invariant and equivariant features, leading to more robust and generalizable visual representations for computer vision tasks.
Abstract:We provide the first method allowing to retrieve spaceborne SIF maps at 30 m ground resolution with a strong correlation ($r^2=0.6$) to high-quality airborne estimates of sun-induced fluorescence (SIF). SIF estimates can provide explanatory information for many tasks related to agricultural management and physiological studies. While SIF products from airborne platforms are accurate and spatially well resolved, the data acquisition of such products remains science-oriented and limited to temporally constrained campaigns. Spaceborne SIF products on the other hand are available globally with often sufficient revisit times. However, the spatial resolution of spaceborne SIF products is too small for agricultural applications. In view of ESA's upcoming FLEX mission we develop a method for SIF retrieval in the O$_2$-A band of hyperspectral DESIS imagery to provide first insights for spaceborne SIF retrieval at high spatial resolution. To this end, we train a simulation-based self-supervised network with a novel perturbation based regularizer and test performance improvements under additional supervised regularization of atmospheric variable prediction. In a validation study with corresponding HyPlant derived SIF estimates at 740 nm we find that our model reaches a mean absolute difference of 0.78 mW / nm / sr / m$^2$.
Abstract:In the acquisition of Magnetic Resonance (MR) images shorter scan times lead to higher image noise. Therefore, automatic image denoising using deep learning methods is of high interest. MR images containing line-like structures such as roots or vessels yield special characteristics as they display connected structures and yield sparse information. For this kind of data, it is important to consider voxel neighborhoods when training a denoising network. In this paper, we translate the Perceptual Loss to 3D data by comparing feature maps of untrained networks in the loss function as done previously for 2D data. We tested the performance of untrained Perceptual Loss (uPL) on 3D image denoising of MR images displaying brain vessels (MR angiograms - MRA) and images of plant roots in soil. We investigate the impact of various uPL characteristics such as weight initialization, network depth, kernel size, and pooling operations on the results. We tested the performance of the uPL loss on four Rician noise levels using evaluation metrics such as the Structural Similarity Index Metric (SSIM). We observe, that our uPL outperforms conventional loss functions such as the L1 loss or a loss based on the Structural Similarity Index Metric (SSIM). The uPL network's initialization is not important, while network depth and pooling operations impact denoising performance. E.g. for both datasets a network with five convolutional layers led to the best performance while a network with more layers led to a performance drop. We also find that small uPL networks led to better or comparable results than using large networks such as VGG. We observe superior performance of our loss for both datasets, all noise levels, and three network architectures. In conclusion, for images containing line-like structures, uPL is an alternative to other loss functions for 3D image denoising.
Abstract:Microfluidic Live-Cell Imaging (MLCI) generates high-quality data that allows biotechnologists to study cellular growth dynamics in detail. However, obtaining these continuous data over extended periods is challenging, particularly in achieving accurate and consistent real-time event classification at the intersection of imaging and stochastic biology. To address this issue, we introduce the Experiment Automation Pipeline for Event-Driven Microscopy to Smart Microfluidic Single-Cells Analysis (EAP4EMSIG). In particular, we present initial zero-shot results from the real-time segmentation module of our approach. Our findings indicate that among four State-Of-The- Art (SOTA) segmentation methods evaluated, Omnipose delivers the highest Panoptic Quality (PQ) score of 0.9336, while Contour Proposal Network (CPN) achieves the fastest inference time of 185 ms with the second-highest PQ score of 0.8575. Furthermore, we observed that the vision foundation model Segment Anything is unsuitable for this particular use case.
Abstract:State-of-the-art computer vision tasks, like monocular depth estimation (MDE), rely heavily on large, modern Transformer-based architectures. However, their application in safety-critical domains demands reliable predictive performance and uncertainty quantification. While Bayesian neural networks provide a conceptually simple approach to serve those requirements, they suffer from the high dimensionality of the parameter space. Parameter-efficient fine-tuning (PEFT) methods, in particular low-rank adaptations (LoRA), have emerged as a popular strategy for adapting large-scale models to down-stream tasks by performing parameter inference on lower-dimensional subspaces. In this work, we investigate the suitability of PEFT methods for subspace Bayesian inference in large-scale Transformer-based vision models. We show that, indeed, combining BitFit, DiffFit, LoRA, and CoLoRA, a novel LoRA-inspired PEFT method, with Bayesian inference enables more robust and reliable predictive performance in MDE.
Abstract:Computer Vision problems deal with the semantic extraction of information from camera images. Especially for field crop images, the underlying problems are hard to label and even harder to learn, and the availability of high-quality training data is low. Deep neural networks do a good job of extracting the necessary models from training examples. However, they rely on an abundance of training data that is not feasible to generate or label by expert annotation. To address this challenge, we make use of the Unreal Engine to render large and complex virtual scenes. We rely on the performance of individual nodes by distributing plant simulations across nodes and both generate scenes as well as train neural networks on GPUs, restricting node communication to parallel learning.