Large Language Models have demonstrated remarkable performance across various tasks, exhibiting the capacity to swiftly acquire new skills, such as through In-Context Learning (ICL) with minimal demonstration examples. In this work, we present a comprehensive framework for investigating Multimodal ICL (M-ICL) in the context of Large Multimodal Models. We consider the best open-source multimodal models (e.g., IDEFICS, OpenFlamingo) and a wide range of multimodal tasks. Our study unveils several noteworthy findings: (1) M-ICL primarily relies on text-driven mechanisms, showing little to no influence from the image modality. (2) When used with advanced-ICL strategy (like RICES), M-ICL is not better than a simple strategy based on majority voting over context examples. Moreover, we identify several biases and limitations of M-ICL that warrant consideration prior to deployment. Code available at https://gitlab.com/folbaeni/multimodal-icl

Autonomous vehicles ought to predict the surrounding agents' trajectories to allow safe maneuvers in uncertain and complex traffic situations. As companies increasingly apply trajectory prediction in the real world, security becomes a relevant concern. In this paper, we focus on backdoors - a security threat acknowledged in other fields but so far overlooked for trajectory prediction. To this end, we describe and investigate four triggers that could affect trajectory prediction. We then show that these triggers (for example, a braking vehicle), when correlated with a desired output (for example, a curve) during training, cause the desired output of a state-of-the-art trajectory prediction model. In other words, the model has good benign performance but is vulnerable to backdoors. This is the case even if the trigger maneuver is performed by a non-casual agent behind the target vehicle. As a side-effect, our analysis reveals interesting limitations within trajectory prediction models. Finally, we evaluate a range of defenses against backdoors. While some, like simple offroad checks, do not enable detection for all triggers, clustering is a promising candidate to support manual inspection to find backdoors.

Large Language Models (LLMs) have made the ambitious quest for generalist agents significantly far from being a fantasy. A key hurdle for building such general models is the diversity and heterogeneity of tasks and modalities. A promising solution is unification, allowing the support of a myriad of tasks and modalities within one unified framework. While few large models (e.g., Flamingo (Alayrac et al., 2022), trained on massive datasets, can support more than two modalities, current small to mid-scale unified models are still limited to 2 modalities, usually image-text or video-text. The question that we ask is: is it possible to build efficiently a unified model that can support all modalities? To answer this, we propose UnIVAL, a step further towards this ambitious goal. Without relying on fancy datasets sizes or models with billions of parameters, the ~ 0.25B parameter UnIVAL model goes beyond two modalities and unifies text, images, video, and audio into a single model. Our model is efficiently pretrained on many tasks, based on task balancing and multimodal curriculum learning. UnIVAL shows competitive performance to existing state-of-the-art approaches, across image and video-text tasks. The feature representations learned from image and video-text modalities, allows the model to achieve competitive performance when finetuned on audio-text tasks, despite not being pretrained on audio. Thanks to the unified model, we propose a novel study on multimodal model merging via weight interpolation of models trained on different multimodal tasks, showing their benefits in particular for out-of-distribution generalization. Finally, we motivate unification by showing the synergy between tasks. The model weights and code are released here: https://github.com/mshukor/UnIVAL.

Assessing the reliability of perception models to covariate shifts and out-of-distribution (OOD) detection is crucial for safety-critical applications such as autonomous vehicles. By nature of the task, however, the relevant data is difficult to collect and annotate. In this paper, we challenge cutting-edge generative models to automatically synthesize data for assessing reliability in semantic segmentation. By fine-tuning Stable Diffusion, we perform zero-shot generation of synthetic data in OOD domains or inpainted with OOD objects. Synthetic data is employed to provide an initial assessment of pretrained segmenters, thereby offering insights into their performance when confronted with real edge cases. Through extensive experiments, we demonstrate a high correlation between the performance on synthetic data and the performance on real OOD data, showing the validity approach. Furthermore, we illustrate how synthetic data can be utilized to enhance the calibration and OOD detection capabilities of segmenters.

Monocular 3D human pose estimation (3D-HPE) is an inherently ambiguous task, as a 2D pose in an image might originate from different possible 3D poses. Yet, most 3D-HPE methods rely on regression models, which assume a one-to-one mapping between inputs and outputs. In this work, we provide theoretical and empirical evidence that, because of this ambiguity, common regression models are bound to predict topologically inconsistent poses, and that traditional evaluation metrics, such as the MPJPE, P-MPJPE and PCK, are insufficient to assess this aspect. As a solution, we propose ManiPose, a novel manifold-constrained multi-hypothesis model capable of proposing multiple candidate 3D poses for each 2D input, together with their corresponding plausibility. Unlike previous multi-hypothesis approaches, our solution is completely supervised and does not rely on complex generative models, thus greatly facilitating its training and usage. Furthermore, by constraining our model to lie within the human pose manifold, we can guarantee the consistency of all hypothetical poses predicted with our approach, which was not possible in previous works. We illustrate the usefulness of ManiPose in a synthetic 1D-to-2D lifting setting and demonstrate on real-world datasets that it outperforms state-of-the-art models in pose consistency by a large margin, while still reaching competitive MPJPE performance.

Large Language Models (LLMs) have so far impressed the world, with unprecedented capabilities that emerge in models at large scales. On the vision side, transformer models (i.e., ViT) are following the same trend, achieving the best performance on challenging benchmarks. With the abundance of such unimodal models, a natural question arises; do we need also to follow this trend to tackle multimodal tasks? In this work, we propose to rather direct effort to efficient adaptations of existing models, and propose to augment Language Models with perception. Existing approaches for adapting pretrained models for vision-language tasks still rely on several key components that hinder their efficiency. In particular, they still train a large number of parameters, rely on large multimodal pretraining, use encoders (e.g., CLIP) trained on huge image-text datasets, and add significant inference overhead. In addition, most of these approaches have focused on Zero-Shot and In Context Learning, with little to no effort on direct finetuning. We investigate the minimal computational effort needed to adapt unimodal models for multimodal tasks and propose a new challenging setup, alongside different approaches, that efficiently adapts unimodal pretrained models. We show that by freezing more than 99% of total parameters, training only one linear projection layer, and prepending only one trainable token, our approach (dubbed eP-ALM) significantly outperforms other baselines on VQA and Captioning across Image, Video, and Audio modalities, following the proposed setup. The code is available here: https://github.com/mshukor/eP-ALM.

Foundation models are first pre-trained on vast unsupervised datasets and then fine-tuned on labeled data. Reinforcement learning, notably from human feedback (RLHF), can further align the network with the intended usage. Yet the imperfections in the proxy reward may hinder the training and lead to suboptimal results; the diversity of objectives in real-world tasks and human opinions exacerbate the issue. This paper proposes embracing the heterogeneity of diverse rewards by following a multi-policy strategy. Rather than focusing on a single a priori reward, we aim for Pareto-optimal generalization across the entire space of preferences. To this end, we propose rewarded soup, first specializing multiple networks independently (one for each proxy reward) and then interpolating their weights linearly. This succeeds empirically because we show that the weights remain linearly connected when fine-tuned on diverse rewards from a shared pre-trained initialization. We demonstrate the effectiveness of our approach for text-to-text (summarization, Q&A, helpful assistant, review), text-image (image captioning, text-to-image generation, visual grounding, VQA), and control (locomotion) tasks. We hope to enhance the alignment of deep models, and how they interact with the world in all its diversity.

Motion forecasting is crucial in enabling autonomous vehicles to anticipate the future trajectories of surrounding agents. To do so, it requires solving mapping, detection, tracking, and then forecasting problems, in a multi-step pipeline. In this complex system, advances in conventional forecasting methods have been made using curated data, i.e., with the assumption of perfect maps, detection, and tracking. This paradigm, however, ignores any errors from upstream modules. Meanwhile, an emerging end-to-end paradigm, that tightly integrates the perception and forecasting architectures into joint training, promises to solve this issue. So far, however, the evaluation protocols between the two methods were incompatible and their comparison was not possible. In fact, and perhaps surprisingly, conventional forecasting methods are usually not trained nor tested in real-world pipelines (e.g., with upstream detection, tracking, and mapping modules). In this work, we aim to bring forecasting models closer to real-world deployment. First, we propose a unified evaluation pipeline for forecasting methods with real-world perception inputs, allowing us to compare the performance of conventional and end-to-end methods for the first time. Second, our in-depth study uncovers a substantial performance gap when transitioning from curated to perception-based data. In particular, we show that this gap (1) stems not only from differences in precision but also from the nature of imperfect inputs provided by perception modules, and that (2) is not trivially reduced by simply finetuning on perception outputs. Based on extensive experiments, we provide recommendations for critical areas that require improvement and guidance towards more robust motion forecasting in the real world. We will release an evaluation library to benchmark models under standardized and practical conditions.

Denoising Diffusion Probabilistic Models (DDPMs) have recently achieved remarkable results in conditional and unconditional image generation. The pre-trained models can be adapted without further training to different downstream tasks, by guiding their iterative denoising process at inference time to satisfy additional constraints. For the specific task of image inpainting, the current guiding mechanism relies on copying-and-pasting the known regions from the input image at each denoising step. However, diffusion models are strongly conditioned by the initial random noise, and therefore struggle to harmonize predictions inside the inpainting mask with the real parts of the input image, often producing results with unnatural artifacts. Our method, dubbed GradPaint, steers the generation towards a globally coherent image. At each step in the denoising process, we leverage the model's "denoised image estimation" by calculating a custom loss measuring its coherence with the masked input image. Our guiding mechanism uses the gradient obtained from backpropagating this loss through the diffusion model itself. GradPaint generalizes well to diffusion models trained on various datasets, improving upon current state-of-the-art supervised and unsupervised methods.

We present an innovative approach to 3D Human Pose Estimation (3D-HPE) by integrating cutting-edge diffusion models, which have revolutionized diverse fields, but are relatively unexplored in 3D-HPE. We show that diffusion models enhance the accuracy, robustness, and coherence of human pose estimations. We introduce DiffHPE, a novel strategy for harnessing diffusion models in 3D-HPE, and demonstrate its ability to refine standard supervised 3D-HPE. We also show how diffusion models lead to more robust estimations in the face of occlusions, and improve the time-coherence and the sagittal symmetry of predictions. Using the Human\,3.6M dataset, we illustrate the effectiveness of our approach and its superiority over existing models, even under adverse situations where the occlusion patterns in training do not match those in inference. Our findings indicate that while standalone diffusion models provide commendable performance, their accuracy is even better in combination with supervised models, opening exciting new avenues for 3D-HPE research.