Learning to Ground VLMs without Forgetting

Spatial awareness is key to enable embodied multimodal AI systems. Yet, without vast amounts of spatial supervision, current Visual Language Models (VLMs) struggle at this task. In this paper, we introduce LynX, a framework that equips pretrained VLMs with visual grounding ability without forgetting their existing image and language understanding skills. To this end, we propose a Dual Mixture of Experts module that modifies only the decoder layer of the language model, using one frozen Mixture of Experts (MoE) pre-trained on image and language understanding and another learnable MoE for new grounding capabilities. This allows the VLM to retain previously learned knowledge and skills, while acquiring what is missing. To train the model effectively, we generate a high-quality synthetic dataset we call SCouT, which mimics human reasoning in visual grounding. This dataset provides rich supervision signals, describing a step-by-step multimodal reasoning process, thereby simplifying the task of visual grounding. We evaluate LynX on several object detection and visual grounding datasets, demonstrating strong performance in object detection, zero-shot localization and grounded reasoning while maintaining its original image and language understanding capabilities on seven standard benchmark datasets.

Revisiting Proposal-based Object Detection

This paper revisits the pipeline for detecting objects in images with proposals. For any object detector, the obtained box proposals or queries need to be classified and regressed towards ground truth boxes. The common solution for the final predictions is to directly maximize the overlap between each proposal and the ground truth box, followed by a winner-takes-all ranking or non-maximum suppression. In this work, we propose a simple yet effective alternative. For proposal regression, we solve a simpler problem where we regress to the area of intersection between proposal and ground truth. In this way, each proposal only specifies which part contains the object, avoiding a blind inpainting problem where proposals need to be regressed beyond their visual scope. In turn, we replace the winner-takes-all strategy and obtain the final prediction by taking the union over the regressed intersections of a proposal group surrounding an object. Our revisited approach comes with minimal changes to the detection pipeline and can be plugged into any existing method. We show that our approach directly improves canonical object detection and instance segmentation architectures, highlighting the utility of intersection-based regression and grouping.

Detecting Objects with Graph Priors and Graph Refinement

The goal of this paper is to detect objects by exploiting their interrelationships. Rather than relying on predefined and labeled graph structures, we infer a graph prior from object co-occurrence statistics. The key idea of our paper is to model object relations as a function of initial class predictions and co-occurrence priors to generate a graph representation of an image for improved classification and bounding box regression. We additionally learn the object-relation joint distribution via energy based modeling. Sampling from this distribution generates a refined graph representation of the image which in turn produces improved detection performance. Experiments on the Visual Genome and MS-COCO datasets demonstrate our method is detector agnostic, end-to-end trainable, and especially beneficial for rare object classes. What is more, we establish a consistent improvement over object detectors like DETR and Faster-RCNN, as well as state-of-the-art methods modeling object interrelationships.

Reinforced Feature Points: Optimizing Feature Detection and Description for a High-Level Task

We address a core problem of computer vision: Detection and description of 2D feature points for image matching. For a long time, hand-crafted designs, like the seminal SIFT algorithm, were unsurpassed in accuracy and efficiency. Recently, learned feature detectors emerged that implement detection and description using neural networks. Training these networks usually resorts to optimizing low-level matching scores, often pre-defining sets of image patches which should or should not match, or which should or should not contain key points. Unfortunately, increased accuracy for these low-level matching scores does not necessarily translate to better performance in high-level vision tasks. We propose a new training methodology which embeds the feature detector in a complete vision pipeline, and where the learnable parameters are trained in an end-to-end fashion. We overcome the discrete nature of key point selection and descriptor matching using principles from reinforcement learning. As an example, we address the task of relative pose estimation between a pair of images. We demonstrate that the accuracy of a state-of-the-art learning-based feature detector can be increased when trained for the task it is supposed to solve at test time. Our training methodology poses little restrictions on the task to learn, and works for any architecture which predicts key point heat maps, and descriptors for key point locations.