Robots' acceptability among humans and their sociability can be significantly enhanced by incorporating human-like reactions. Humans can react to environmental events very quickly and without thinking. An instance where humans display natural reactions is when they encounter a sudden and loud sound that startles or frightens them. During such moments, individuals may instinctively move their hands, turn toward the origin of the sound, and try to determine the event's cause. This inherent behavior motivated us to explore this less-studied part of social robotics. In this work, a multi-modal system composed of an action generator, sound classifier, and YOLO object detector was designed to sense the environment and, in the presence of sudden loud sounds, show natural human fear reactions, and finally, locate the fear-causing sound source in the environment. These unique and valid generated motions and inferences could imitate intrinsic human reactions and enhance the sociability of robots. For motion generation, a model based on LSTM and MDN networks was proposed to synthesize various motions. Also, in the case of sound detection, a transfer learning model was preferred that used the spectrogram of sound signals as its input. After developing individual models for sound detection, motion generation, and image recognition, they were integrated into a comprehensive fear module that was implemented on the NAO robot. Finally, the fear module was tested in practical application and two groups of experts and non-experts filled out a questionnaire to evaluate the performance of the robot. Given our promising results, this preliminary exploratory research provides a fresh perspective on social robotics and could be a starting point for modeling intrinsic human behaviors and emotions in robots.
SCONE-GAN presents an end-to-end image translation, which is shown to be effective for learning to generate realistic and diverse scenery images. Most current image-to-image translation approaches are devised as two mappings: a translation from the source to target domain and another to represent its inverse. While successful in many applications, these approaches may suffer from generating trivial solutions with limited diversity. That is because these methods learn more frequent associations rather than the scene structures. To mitigate the problem, we propose SCONE-GAN that utilises graph convolutional networks to learn the objects dependencies, maintain the image structure and preserve its semantics while transferring images into the target domain. For more realistic and diverse image generation we introduce style reference image. We enforce the model to maximize the mutual information between the style image and output. The proposed method explicitly maximizes the mutual information between the related patches, thus encouraging the generator to produce more diverse images. We validate the proposed algorithm for image-to-image translation and stylizing outdoor images. Both qualitative and quantitative results demonstrate the effectiveness of our approach on four dataset.
With the prevalence of the Pretraining-Finetuning paradigm in transfer learning, the robustness of downstream tasks has become a critical concern. In this work, we delve into adversarial robustness in transfer learning and reveal the critical role of initialization, including both the pretrained model and the linear head. First, we discover the necessity of an adversarially robust pretrained model. Specifically, we reveal that with a standard pretrained model, Parameter-Efficient Finetuning~(PEFT) methods either fail to be adversarially robust or continue to exhibit significantly degraded adversarial robustness on downstream tasks, even with adversarial training during finetuning. Leveraging a robust pretrained model, surprisingly, we observe that a simple linear probing can outperform full finetuning and other PEFT methods with random initialization on certain datasets. We further identify that linear probing excels in preserving robustness from the robust pretraining. Based on this, we propose Robust Linear Initialization~(RoLI) for adversarial finetuning, which initializes the linear head with the weights obtained by adversarial linear probing to maximally inherit the robustness from pretraining. Across five different image classification datasets, we demonstrate the effectiveness of RoLI and achieve new state-of-the-art results.
One of the motivations for explainable AI is to allow humans to make better and more informed decisions regarding the use and deployment of AI models. But careful evaluations are needed to assess whether this expectation has been fulfilled. Current evaluations mainly focus on algorithmic properties of explanations, and those that involve human subjects often employ subjective questions to test human's perception of explanation usefulness, without being grounded in objective metrics and measurements. In this work, we evaluate whether explanations can improve human decision-making in practical scenarios of machine learning model development. We conduct a mixed-methods user study involving image data to evaluate saliency maps generated by SmoothGrad, GradCAM, and an oracle explanation on two tasks: model selection and counterfactual simulation. To our surprise, we did not find evidence of significant improvement on these tasks when users were provided with any of the saliency maps, even the synthetic oracle explanation designed to be simple to understand and highly indicative of the answer. Nonetheless, explanations did help users more accurately describe the models. These findings suggest caution regarding the usefulness and potential for misunderstanding in saliency-based explanations.
In sufficiently complex tasks, it is expected that as a side effect of learning to solve a problem, a neural network will learn relevant abstractions of the representation of that problem. This has been confirmed in particular in machine vision where a number of works showed that correlations could be found between the activations of specific units (neurons) in a neural network and the visual concepts (textures, colors, objects) present in the image. Here, we explore the use of self-organizing maps as a way to both visually and computationally inspect how activation vectors of whole layers of neural networks correspond to neural representations of abstract concepts such as `female person' or `realist painter'. We experiment with multiple measures applied to those maps to assess the level of representation of a concept in a network's layer. We show that, among the measures tested, the relative entropy of the activation map for a concept compared to the map for the whole data is a suitable candidate and can be used as part of a methodology to identify and locate the neural representation of a concept, visualize it, and understand its importance in solving the prediction task at hand.
Neural radiance fields (NeRF) is a promising approach for generating photorealistic images and representing complex scenes. However, when processing data sequentially, it can suffer from catastrophic forgetting, where previous data is easily forgotten after training with new data. Existing incremental learning methods using knowledge distillation assume that continuous data chunks contain both 2D images and corresponding camera pose parameters, pre-estimated from the complete dataset. This poses a paradox as the necessary camera pose must be estimated from the entire dataset, even though the data arrives sequentially and future chunks are inaccessible. In contrast, we focus on a practical scenario where camera poses are unknown. We propose IL-NeRF, a novel framework for incremental NeRF training, to address this challenge. IL-NeRF's key idea lies in selecting a set of past camera poses as references to initialize and align the camera poses of incoming image data. This is followed by a joint optimization of camera poses and replay-based NeRF distillation. Our experiments on real-world indoor and outdoor scenes show that IL-NeRF handles incremental NeRF training and outperforms the baselines by up to $54.04\%$ in rendering quality.
While Multi-modal Language Models (MLMs) demonstrate impressive multimodal ability, they still struggle on providing factual and precise responses for tasks like visual question answering (VQA). In this paper, we address this challenge from the perspective of contextual information. We propose Causal Context Generation, Causal-CoG, which is a prompting strategy that engages contextual information to enhance precise VQA during inference. Specifically, we prompt MLMs to generate contexts, i.e, text description of an image, and engage the generated contexts for question answering. Moreover, we investigate the advantage of contexts on VQA from a causality perspective, introducing causality filtering to select samples for which contextual information is helpful. To show the effectiveness of Causal-CoG, we run extensive experiments on 10 multimodal benchmarks and show consistent improvements, e.g., +6.30% on POPE, +13.69% on Vizwiz and +6.43% on VQAv2 compared to direct decoding, surpassing existing methods. We hope Casual-CoG inspires explorations of context knowledge in multimodal models, and serves as a plug-and-play strategy for MLM decoding.
In unstructured outdoor environments, robotics requires accurate and efficient odometry with low computational time. Existing low-bias LiDAR odometry methods are often computationally expensive. To address this problem, we present a lightweight LiDAR odometry method that converts unorganized point cloud data into a spherical range image (SRI) and filters out surface, edge, and ground features in the image plane. This substantially reduces computation time and the required features for odometry estimation in LOAM-based algorithms. Our odometry estimation method does not rely on global maps or loop closure algorithms, which further reduces computational costs. Experimental results generate a translation and rotation error of 0.86\% and 0.0036{\deg}/m on the KITTI dataset with an average runtime of 78ms. In addition, we tested the method with our data, obtaining an average closed-loop error of 0.8m and a runtime of 27ms over eight loops covering 3.5Km.
The Segment Anything Model (SAM) has achieved a notable success in two-dimensional image segmentation in natural images. However, the substantial gap between medical and natural images hinders its direct application to medical image segmentation tasks. Particularly in 3D medical images, SAM struggles to learn contextual relationships between slices, limiting its practical applicability. Moreover, applying 2D SAM to 3D images requires prompting the entire volume, which is time- and label-consuming. To address these problems, we propose Slide-SAM, which treats a stack of three adjacent slices as a prediction window. It firstly takes three slices from a 3D volume and point- or bounding box prompts on the central slice as inputs to predict segmentation masks for all three slices. Subsequently, the masks of the top and bottom slices are then used to generate new prompts for adjacent slices. Finally, step-wise prediction can be achieved by sliding the prediction window forward or backward through the entire volume. Our model is trained on multiple public and private medical datasets and demonstrates its effectiveness through extensive 3D segmetnation experiments, with the help of minimal prompts. Code is available at \url{https://github.com/Curli-quan/Slide-SAM}.
Enhancing a low-light noisy RAW image into a well-exposed and clean sRGB image is a significant challenge in computational photography. Due to the limitation of large-scale paired data, prior approaches have difficulty in recovering fine details and true colors in extremely low-light regions. Meanwhile, recent advancements in generative diffusion models have shown promising generating capabilities, which inspires this work to explore generative priors from a diffusion model trained on a large-scale open-domain dataset to benefit the low-light image enhancement (LLIE) task. Based on this intention, we propose a novel diffusion-model-based LLIE method, dubbed LDM-SID. LDM-SID aims at inserting a set of proposed taming modules into a frozen pre-trained diffusion model to steer its generating process. Specifically, the taming module fed with low-light information serves to output a pair of affine transformation parameters to modulate the intermediate feature in the diffusion model. Additionally, based on the observation of dedicated generative priors across different portions of the diffusion model, we propose to apply 2D discrete wavelet transforms on the input RAW image, resulting in dividing the LLIE task into two essential parts: low-frequency content generation and high-frequency detail maintenance. This enables us to skillfully tame the diffusion model for optimized structural generation and detail enhancement. Extensive experiments demonstrate the proposed method not only achieves state-of-the-art performance in quantitative evaluations but also shows significant superiority in visual comparisons. These findings highlight the effectiveness of leveraging a pre-trained diffusion model as a generative prior to the LLIE task.