Gertjan Burghouts

Valentinos Pariza, Mohammadreza Salehi, Gertjan Burghouts et al.

We introduce NeCo: Patch Neighbor Consistency, a novel self-supervised training loss that enforces patch-level nearest neighbor consistency across a student and teacher model. Compared to contrastive approaches that only yield binary learning signals, i.e., 'attract' and 'repel', this approach benefits from the more fine-grained learning signal of sorting spatially dense features relative to reference patches. Our method leverages differentiable sorting applied on top of pretrained representations, such as DINOv2-registers to bootstrap the learning signal and further improve upon them. This dense post-pretraining leads to superior performance across various models and datasets, despite requiring only 19 hours on a single GPU. This method generates high-quality dense feature encoders and establishes several new state-of-the-art results such as +5.5% and +6% for non-parametric in-context semantic segmentation on ADE20k and Pascal VOC, +7.2% and +5.7% for linear segmentation evaluations on COCO-Things and -Stuff and improvements in the 3D understanding of multi-view consistency on SPair-71k, by more than 1.5%.

Fieke Hillerstrom, Gertjan Burghouts

Many inductive logic programming (ILP) methods are incapable of learning programs from probabilistic background knowledge, e.g. coming from sensory data or neural networks with probabilities. We propose Propper, which handles flawed and probabilistic background knowledge by extending ILP with a combination of neurosymbolic inference, a continuous criterion for hypothesis selection (BCE) and a relaxation of the hypothesis constrainer (NoisyCombo). For relational patterns in noisy images, Propper can learn programs from as few as 8 examples. It outperforms binary ILP and statistical models such as a Graph Neural Network.

Gertjan Burghouts, Fieke Hillerström, Erwin Walraven et al.

We consider the problem of finding spatial configurations of multiple objects in images, e.g., a mobile inspection robot is tasked to localize abandoned tools on the floor. We define the spatial configuration of objects by first-order logic in terms of relations and attributes. A neuro-symbolic program matches the logic formulas to probabilistic object proposals for the given image, provided by language-vision models by querying them for the symbols. This work is the first to combine neuro-symbolic programming (reasoning) and language-vision models (learning) to find spatial configurations of objects in images in an open world setting. We show the effectiveness by finding abandoned tools on floors and leaking pipes. We find that most prediction errors are due to biases in the language-vision model.

Youssef Zahran, Gertjan Burghouts, Yke Bauke Eisma

Despite the significant advancements in computer vision models, their ability to generalize to novel object-attribute compositions remains limited. Existing methods for Compositional Zero-Shot Learning (CZSL) mainly focus on image classification. This paper aims to enhance CZSL in object detection without forgetting prior learned knowledge. We use Grounding DINO and incorporate Compositional Soft Prompting (CSP) into it and extend it with Compositional Anticipation. We achieve a 70.5% improvement over CSP on the harmonic mean (HM) between seen and unseen compositions on the CLEVR dataset. Furthermore, we introduce Contrastive Prompt Tuning to incrementally address model confusion between similar compositions. We demonstrate the effectiveness of this method and achieve an increase of 14.5% in HM across the pretrain, increment, and unseen sets. Collectively, these methods provide a framework for learning various compositions with limited data, as well as improving the performance of underperforming compositions when additional data becomes available.

Emanuele Mezzi, Gertjan Burghouts, Maarten Kruithof

Text-to-image retrieval in remote sensing (RS) has advanced rapidly with the rise of large vision-language models (LVLMs) tailored for aerial and satellite imagery, culminating in remote sensing large vision-language models (RS-LVLMS). However, limited explainability and poor handling of complex spatial relations remain key challenges for real-world use. To address these issues, we introduce RUNE (Reasoning Using Neurosymbolic Entities), an approach that combines Large Language Models (LLMs) with neurosymbolic AI to retrieve images by reasoning over the compatibility between detected entities and First-Order Logic (FOL) expressions derived from text queries. Unlike RS-LVLMs that rely on implicit joint embeddings, RUNE performs explicit reasoning, enhancing performance and interpretability. For scalability, we propose a logic decomposition strategy that operates on conditioned subsets of detected entities, guaranteeing shorter execution time compared to neural approaches. Rather than using foundation models for end-to-end retrieval, we leverage them only to generate FOL expressions, delegating reasoning to a neurosymbolic inference module. For evaluation we repurpose the DOTA dataset, originally designed for object detection, by augmenting it with more complex queries than in existing benchmarks. We show the LLM's effectiveness in text-to-logic translation and compare RUNE with state-of-the-art RS-LVLMs, demonstrating superior performance. We introduce two metrics, Retrieval Robustness to Query Complexity (RRQC) and Retrieval Robustness to Image Uncertainty (RRIU), which evaluate performance relative to query complexity and image uncertainty. RUNE outperforms joint-embedding models in complex RS retrieval tasks, offering gains in performance, robustness, and explainability. We show RUNE's potential for real-world RS applications through a use case on post-flood satellite image retrieval.

Judith Dijk, Gertjan Burghouts, Kapil D. Katyal et al.

For navigation of robots, image segmentation is an important component to determining a terrain's traversability. For safe and efficient navigation, it is key to assess the uncertainty of the predicted segments. Current uncertainty estimation methods are limited to a specific choice of model architecture, are costly in terms of training time, require large memory for inference (ensembles), or involve complex model architectures (energy-based, hyperbolic, masking). In this paper, we propose a simple, light-weight module that can be connected to any pretrained image segmentation model, regardless of its architecture, with marginal additional computation cost because it reuses the model's backbone. Our module is based on maximum separation of the segmentation classes by respective prototype vectors. This optimizes the probability that out-of-distribution segments are projected in between the prototype vectors. The uncertainty value in the classification label is obtained from the distance to the nearest prototype. We demonstrate the effectiveness of our module for terrain segmentation.

Anne Kemmeren, Gertjan Burghouts, Michael van Bekkum et al.

For effective interactions with the open world, robots should understand how interactions with known and novel objects help them towards their goal. A key aspect of this understanding lies in detecting an object's affordances, which represent the potential effects that can be achieved by manipulating the object in various ways. Our approach leverages a dialogue of large language models (LLMs) and vision-language models (VLMs) to achieve open-world affordance detection. Given open-vocabulary descriptions of intended actions and effects, the useful objects in the environment are found. By grounding our system in the physical world, we account for the robot's embodiment and the intrinsic properties of the objects it encounters. In our experiments, we have shown that our method produces tailored outputs based on different embodiments or intended effects. The method was able to select a useful object from a set of distractors. Finetuning the VLM for physical properties improved overall performance. These results underline the importance of grounding the affordance search in the physical world, by taking into account robot embodiment and the physical properties of objects.

Gertjan Burghouts, Marianne Schaaphok, Michael van Bekkum et al.

Mobile robot platforms will increasingly be tasked with activities that involve grasping and manipulating objects in open world environments. Affordance understanding provides a robot with means to realise its goals and execute its tasks, e.g. to achieve autonomous navigation in unknown buildings where it has to find doors and ways to open these. In order to get actionable suggestions, robots need to be able to distinguish subtle differences between objects, as they may result in different action sequences: doorknobs require grasp and twist, while handlebars require grasp and push. In this paper, we improve affordance perception for a robot in an open-world setting. Our contribution is threefold: (1) We provide an affordance representation with precise, actionable affordances; (2) We connect this knowledge base to a foundational vision-language models (VLM) and prompt the VLM for a wider variety of new and unseen objects; (3) We apply a human-in-the-loop for corrections on the output of the VLM. The mix of affordance representation, image detection and a human-in-the-loop is effective for a robot to search for objects to achieve its goals. We have demonstrated this in a scenario of finding various doors and the many different ways to open them.

Michail Tsiaousis, Gertjan Burghouts, Fieke Hillerström et al.

The dominant paradigm in spatiotemporal action detection is to classify actions using spatiotemporal features learned by 2D or 3D Convolutional Networks. We argue that several actions are characterized by their context, such as relevant objects and actors present in the video. To this end, we introduce an architecture based on self-attention and Graph Convolutional Networks in order to model contextual cues, such as actor-actor and actor-object interactions, to improve human action detection in video. We are interested in achieving this in a weakly-supervised setting, i.e. using as less annotations as possible in terms of action bounding boxes. Our model aids explainability by visualizing the learned context as an attention map, even for actions and objects unseen during training. We evaluate how well our model highlights the relevant context by introducing a quantitative metric based on recall of objects retrieved by attention maps. Our model relies on a 3D convolutional RGB stream, and does not require expensive optical flow computation. We evaluate our models on the DALY dataset, which consists of human-object interaction actions. Experimental results show that our contextualized approach outperforms a baseline action detection approach by more than 2 points in Video-mAP. Code is available at \url{https://github.com/micts/acgcn}

Dennis Hoftijzer, Gertjan Burghouts, Luuk Spreeuwers

Deep Reinforcement Learning (DRL) has shown great potential in enabling robots to find certain objects (e.g., `find a fridge') in environments like homes or schools. This task is known as Object-Goal Navigation (ObjectNav). DRL methods are predominantly trained and evaluated using environment simulators. Although DRL has shown impressive results, the simulators may be biased or limited. This creates a risk of shortcut learning, i.e., learning a policy tailored to specific visual details of training environments. We aim to deepen our understanding of shortcut learning in ObjectNav, its implications and propose a solution. We design an experiment for inserting a shortcut bias in the appearance of training environments. As a proof-of-concept, we associate room types to specific wall colors (e.g., bedrooms with green walls), and observe poor generalization of a state-of-the-art (SOTA) ObjectNav method to environments where this is not the case (e.g., bedrooms with blue walls). We find that shortcut learning is the root cause: the agent learns to navigate to target objects, by simply searching for the associated wall color of the target object's room. To solve this, we propose Language-Based (L-B) augmentation. Our key insight is that we can leverage the multimodal feature space of a Vision-Language Model (VLM) to augment visual representations directly at the feature-level, requiring no changes to the simulator, and only an addition of one layer to the model. Where the SOTA ObjectNav method's success rate drops 69%, our proposal has only a drop of 23%.

George Potter, Gertjan Burghouts, Joris Sijs

Affordances enable robots to have a semantic understanding of their surroundings. This allows them to have more acting flexibility when completing a given task. Capturing object affordances in a machine learning model is a difficult task, because of their dependence on contextual information. Markov Logic Networks (MLN) combine probabilistic reasoning with logic that is able to capture such context. Mobile robots operate in partially known environments wherein unseen object affordances can be observed. This new information must be incorporated into the existing knowledge, without having to retrain the MLN from scratch. We introduce the MLN Cumulative Learning Algorithm (MLN-CLA). MLN-CLA learns new relations in various knowledge domains by retaining knowledge and only updating the changed knowledge, for which the MLN is retrained. We show that MLN-CLA is effective for accumulative learning and zero-shot affordance inference, outperforming strong baselines.

Jan Erik van Woerden, Gertjan Burghouts, Lotte Nijskens et al.

Vision-language models (VLMs) like CLIP enable zero-shot classification by aligning images and text in a shared embedding space, offering advantages for defense applications with scarce labeled data. However, CLIP's robustness in challenging military environments, with partial occlusion and degraded signal-to-noise ratio (SNR), remains underexplored. We investigate CLIP variants' robustness to occlusion using a custom dataset of 18 military vehicle classes and evaluate using Normalized Area Under the Curve (NAUC) across occlusion percentages. Four key insights emerge: (1) Transformer-based CLIP models consistently outperform CNNs, (2) fine-grained, dispersed occlusions degrade performance more than larger contiguous occlusions, (3) despite improved accuracy, performance of linear-probed models sharply drops at around 35% occlusion, (4) by finetuning the model's backbone, this performance drop occurs at more than 60% occlusion. These results underscore the importance of occlusion-specific augmentations during training and the need for further exploration into patch-level sensitivity and architectural resilience for real-world deployment of CLIP.

Frank Ruis, Gertjan Burghouts, Hugo Kuijf

Recent progress in large pre-trained vision language models (VLMs) has reached state-of-the-art performance on several object detection benchmarks and boasts strong zero-shot capabilities, but for optimal performance on specific targets some form of finetuning is still necessary. While the initial VLM weights allow for great few-shot transfer learning, this usually involves the loss of the original natural language querying and zero-shot capabilities. Inspired by the success of Textual Inversion (TI) in personalizing text-to-image diffusion models, we propose a similar formulation for open-vocabulary object detection. TI allows extending the VLM vocabulary by learning new or improving existing tokens to accurately detect novel or fine-grained objects from as little as three examples. The learned tokens are completely compatible with the original VLM weights while keeping them frozen, retaining the original model's benchmark performance, and leveraging its existing capabilities such as zero-shot domain transfer (e.g., detecting a sketch of an object after training only on real photos). The storage and gradient calculations are limited to the token embedding dimension, requiring significantly less compute than full-model fine-tuning. We evaluated whether the method matches or outperforms the baseline methods that suffer from forgetting in a wide variety of quantitative and qualitative experiments.

Fedor Taggenbrock, Gertjan Burghouts, Ronald Poppe

Matching objects across partially overlapping camera views is crucial in multi-camera systems and requires a view-invariant feature extraction network. Training such a network with cycle-consistency circumvents the need for labor-intensive labeling. In this paper, we extend the mathematical formulation of cycle-consistency to handle partial overlap. We then introduce a pseudo-mask which directs the training loss to take partial overlap into account. We additionally present several new cycle variants that complement each other and present a time-divergent scene sampling scheme that improves the data input for this self-supervised setting. Cross-camera matching experiments on the challenging DIVOTrack dataset show the merits of our approach. Compared to the self-supervised state-of-the-art, we achieve a 4.3 percentage point higher F1 score with our combined contributions. Our improvements are robust to reduced overlap in the training data, with substantial improvements in challenging scenes that need to make few matches between many people. Self-supervised feature networks trained with our method are effective at matching objects in a range of multi-camera settings, providing opportunities for complex tasks like large-scale multi-camera scene understanding.

Eric Brouwer, Jan Erik van Woerden, Gertjan Burghouts et al.

Few-shot, fine-grained classification in computer vision poses significant challenges due to the need to differentiate subtle class distinctions with limited data. This paper presents a novel method that enhances the Contrastive Language-Image Pre-Training (CLIP) model through adaptive prompt tuning, guided by real-time visual inputs. Unlike existing techniques such as Context Optimization (CoOp) and Visual Prompt Tuning (VPT), which are constrained by static prompts or visual token reliance, the proposed approach leverages a cross-attention mechanism to dynamically refine text prompts for the image at hand. This enables an image-specific alignment of textual features with image patches extracted from the Vision Transformer, making the model more effective for datasets with high intra-class variance and low inter-class differences. The method is evaluated on several datasets, including CUBirds, Oxford Flowers, and FGVC Aircraft, showing significant performance gains over static prompt tuning approaches. To ensure these performance gains translate into trustworthy predictions, we integrate Monte-Carlo Dropout in our approach to improve the reliability of the model predictions and uncertainty estimates. This integration provides valuable insights into the model's predictive confidence, helping to identify when predictions can be trusted and when additional verification is necessary. This dynamic approach offers a robust solution, advancing the state-of-the-art for few-shot fine-grained classification.

Leander van den Heuvel, Gertjan Burghouts, David W. Zhang et al.

For object detection, it is possible to view the prediction of bounding boxes as a reverse diffusion process. Using a diffusion model, the random bounding boxes are iteratively refined in a denoising step, conditioned on the image. We propose a stochastic accumulator function that starts each run with random bounding boxes and combines the slightly different predictions. We empirically verify that this improves detection performance. The improved detections are leveraged on unlabelled images as weighted pseudo-labels for semi-supervised learning. We evaluate the method on a challenging out-of-domain test set. Our method brings significant improvements and is on par with human-selected pseudo-labels, while not requiring any human involvement.

Frank Ruis, Gertjan Burghouts, Doina Bucur

Humans are good at compositional zero-shot reasoning; someone who has never seen a zebra before could nevertheless recognize one when we tell them it looks like a horse with black and white stripes. Machine learning systems, on the other hand, usually leverage spurious correlations in the training data, and while such correlations can help recognize objects in context, they hurt generalization. To be able to deal with underspecified datasets while still leveraging contextual clues during classification, we propose ProtoProp, a novel prototype propagation graph method. First we learn prototypical representations of objects (e.g., zebra) that are conditionally independent w.r.t. their attribute labels (e.g., stripes) and vice versa. Next we propagate the independent prototypes through a compositional graph, to learn compositional prototypes of novel attribute-object combinations that reflect the dependencies of the target distribution. The method does not rely on any external data, such as class hierarchy graphs or pretrained word embeddings. We evaluate our approach on AO-Clever, a synthetic and strongly visual dataset with clean labels, and UT-Zappos, a noisy real-world dataset of fine-grained shoe types. We show that in the generalized compositional zero-shot setting we outperform state-of-the-art results, and through ablations we show the importance of each part of the method and their contribution to the final results.