Vision-language models (VLMs) like CLIP have been cherished for their ability to perform zero-shot visual recognition on open-vocabulary concepts. This is achieved by selecting the object category whose textual representation bears the highest similarity with the query image. While successful in some domains, this method struggles with identifying fine-grained entities as well as generalizing to unseen concepts that are not captured by the training distribution. Recent works attempt to mitigate these challenges by integrating category descriptions at test time, albeit yielding modest improvements. We attribute these limited gains to a misalignment between image regions and textual descriptions, which stems from CLIP’s global alignment objective. In this paper, we propose GRAIN, a new pretraining strategy aimed at aligning representations at both fine and coarse levels simultaneously. Our approach learns to jointly ground textual descriptions in image regions along with aligning overarching captions with global image representations. To drive this pre-training, we leverage frozen Multimodal Large Language Models (MLLMs) to derive large-scale synthetic annotations. We demonstrate the enhanced zero-shot performance of our model compared to current state-of-the art methods across 11 diverse image classification datasets. Additionally, we introduce Products-2023, a newly curated, manually labeled dataset featuring novel concepts, and showcase our model’s ability to recognize these concepts by benchmarking on it. Significant improvements achieved by our model on other downstream tasks like retrieval further highlight the superior quality of representations learned by our approach. Code available at https://github.com/shaunak27/grain-clip.

Robust fine-tuning aims to adapt large foundation models to downstream tasks while preserving their robustness to distribution shifts. Existing methods primarily focus on constraining and projecting current model towards the pre-trained initialization based on the magnitudes between fine-tuned and pre-trained weights, which often require extensive hyper-parameter tuning and can sometimes result in underfitting. In this work, we propose $\textbf{Di}$rectional $\textbf{Gra}$dient $\textbf{P}$rojection (DiGraP), a novel layer-wise trainable method that incorporates directional information from gradients to bridge regularization and multi-objective optimization. Besides demonstrating our method on image classification, as another contribution we generalize this area to the multi-modal evaluation settings for robust fine-tuning. Specifically, we first bridge the uni-modal and multi-modal gap by performing analysis on Image Classification reformulated Visual Question Answering (VQA) benchmarks and further categorize ten out-of-distribution (OOD) VQA datasets by distribution shift types and degree (i.e. near versus far OOD). Experimental results show that DiGraP consistently outperforms existing baselines across Image Classfication and VQA tasks with discriminative and generative backbones, improving both in-distribution (ID) generalization and OOD robustness.

Producing quality segmentation masks for images is a fundamental problem in computer vision. Recent research has explored large-scale supervised training to enable zero- shot transfer segmentation on virtually any image style and unsupervised training to enable segmentation without dense annotations. However, constructing a model capable of segmenting anything in a zero-shot manner without any anno-tations is still challenging. In this paper, we propose to uti-lize the self-attention layers in stable diffusion models to achieve this goal because the pre-trained stable diffusion model has learned inherent concepts of objects within its attention layers. Specifically, we introduce a simple yet ef-fective iterative merging process based on measuring KL divergence among attention maps to merge them into valid segmentation masks. The proposed method does not re-quire any training or language dependency to extract qual-ity segmentation for any images. On COCO-Stuff-27, our method surpasses the prior unsupervised zero-shot trans-fer SOTA method by an absolute 26% in pixel accuracy and 17% in mean IoU. The project page is at https://sites.google.com/view/diffseg/home.11Georgia Institute of Technology

Modern optimizers such as AdamW, equipped with momentum and adaptive learning rate, are designed to escape local minima and explore the vast parameter space. This exploration is beneficial for finding good loss basins when training from scratch. It is not necessarily ideal when resuming from a powerful foundation model because it can lead to large deviations from the pre-trained initialization and, consequently, worse robustness and generalization. At the same time, strong regularization on all parameters can lead to under-fitting. We hypothesize that selectively regularizing the parameter space is the key to fitting and retraining the pre-trained knowledge. This paper proposes a new weight decay technique, Selective Projection Decay (SPD), that selectively imposes a strong penalty on certain layers while allowing others to change freely. Intuitively, SPD expands and contracts the parameter search space for layers with consistent and inconsistent loss reduction, respectively. Experimentally, when equipped with SPD, Adam consistently provides better in-distribution generalization and out-of-distribution robustness performance on multiple popular vision and language benchmarks.

Recent studies on transfer learning have shown that selectively fine-tuning a subset of layers or customizing different learning rates for each layer can greatly improve robustness to out-of-distribution (OOD) data and retain generalization capability in the pre-trained models. However, most of these methods employ manually crafted heuristics or expensive hyper-parameter searches, which prevent them from scaling up to large datasets and neural networks. To solve this problem, we propose Trainable Projected Gradient Method (TPGM) to automatically learn the constraint imposed for each layer for a fine-grained fine-tuning regularization. This is motivated by formulating fine-tuning as a bi-level constrained optimization problem. Specifically, TPGM maintains a set of projection radii, i.e., distance constraints between the fine-tuned model and the pretrained model, for each layer, and enforces them through weight projections. To learn the constraints, we propose a bi-level optimization to automatically learn the best set of projection radii in an end-to-end manner. Theoretically, we show that the bi-level optimization formulation is the key to learning different constraints for each layer. Empirically, with little hyper-parameter search cost, TPGM outperforms existing fine-tuning methods in OOD performance while matching the best in-distribution (ID) performance. For example, when fine-tuned on DomainNet-Real and ImageNet, compared to vanilla fine-tuning, TPGM shows 22% and 10% relative OOD improvement respectively on their sketch counterparts. Code is available at https://github.com/PotatoTian/TPGM.

Robust fine-tuning aims to achieve competitive in-distribution (ID) performance while maintaining the out-of-distribution (OOD) robustness of a pre-trained model when transferring it to a downstream task. Recently, projected gradient descent has been successfully used in robust fine-tuning by constraining the deviation from the initialization of the fine-tuned model explicitly through projection. However, algorithmically, two limitations prevent this method from being adopted more widely, scalability and efficiency. In this paper, we propose a new projection-based fine-tuning algorithm, Fast Trainable Projection (FTP) for computationally efficient learning of per-layer projection constraints, resulting in an average 35% speedup on our benchmarks compared to prior works. FTP can be combined with existing optimizers such as AdamW, and be used in a plug-and-play fashion. Finally, we show that FTP is a special instance of hyper-optimizers that tune the hyper-parameters of optimizers in a learnable manner through nested differentiation. Empirically, we show superior robustness on OOD datasets, including domain shifts and natural corruptions, across four different vision tasks with five different pre-trained models. Additionally, we demonstrate that FTP is broadly applicable and beneficial to other learning scenarios such as low-label and continual learning settings thanks to its easy adaptability. The code will be available at https://github.com/GT-RIPL/FTP.git.

Class imbalance is a fundamental problem in computer vision applications such as semantic segmentation. Specifically, uneven class distributions in a training dataset often result in unsatisfactory performance on under-represented classes. Many works have proposed to weight the standard cross entropy loss function with pre-computed weights based on class statistics, such as the number of samples and class margins. There are two major drawbacks to these methods: 1) constantly up-weighting minority classes can introduce excessive false positives in semantic segmentation; 2) a minority class is not necessarily a hard class. The consequence is low precision due to excessive false positives. In this regard, we propose a hard-class mining loss by reshaping the vanilla cross entropy loss such that it weights the loss for each class dynamically based on instantaneous recall performance. We show that the novel recall loss changes gradually between the standard cross entropy loss and the inverse frequency weighted loss. Recall loss also leads to improved mean accuracy while offering competitive mean Intersection over Union (IoU) performance. On Synthia dataset11Synthia-Sequence Summer split, recall loss achieves 9% relative improvement on mean accuracy with competitive mean IoU using DeepLab-ResNet18 compared to the cross entropy loss. Code available at https://github.com/PotatoTian/recall-semseg.

Adapting large-scale pretrained models to various downstream tasks via fine-tuning is a standard method in machine learning. Recently, parameter-efficient fine-tuning methods have shown promise in adapting a pretrained model to different tasks while training only a few parameters. Despite their success, most existing methods are proposed in Natural Language Processing tasks with language Transformers, and adaptation to Computer Vision tasks with Vision Transformers remains under-explored, especially for dense vision tasks. Further, in multi-task settings, individually fine-tuning and storing separate models for different tasks is inefficient. In this work, we provide an extensive single- and multi-task parameter-efficient benchmark and examine existing parameter-efficient fine-tuning NLP methods for vision tasks. Our results on four different dense vision tasks showed that existing methods cannot be efficiently integrated due to the hierarchical nature of the Hierarchical Vision Transformers. To overcome this issue, we propose Polyhistor and Polyhistor-Lite, consisting of Decomposed HyperNetworks and Layer-wise Scaling Kernels, to share information across different tasks with a few trainable parameters. This leads to favorable performance improvements against existing parameter-efficient methods while using fewer trainable parameters. Specifically, Polyhistor achieves competitive accuracy compared to the state-of-the-art while only using less than 10% of their trainable parameters. Furthermore, our methods show larger performance gains when large networks and more pretraining data are used.

In this paper, we address bandwidth-limited and obstruction-prone collaborative perception, specifically in the context of multi-agent semantic segmentation. This setting presents several key challenges, including processing and ex-changing unregistered robotic swarm imagery. To be successful, solutions must effectively leverage multiple non-static and intermittently-overlapping RGB perspectives, while heeding bandwidth constraints and overcoming unwanted foreground obstructions. As such, we propose an end-to-end learn-able Multi-Agent Spatial Handshaking network (MASH) to process, compress, and propagate visual information across a robotic swarm. Our distributed communication module operates directly (and exclusively) on raw image data, without additional input requirements such as pose, depth, or warping data. We demonstrate superior performance of our model compared against several baselines in a photo-realistic multi-robot AirSim environment, especially in the presence of image occlusions. Our method achieves an absolute 11% IoU improvement over strong baselines.

It is well known that vision classification models suffer from poor calibration in the face of data distribution shifts. In this paper, we take a geometric approach to this problem. We propose Geometric Sensitivity Decomposition (GSD) which decomposes the norm of a sample feature embedding and the angular similarity to a target classifier into an instance-dependent and an instance-independent com-ponent. The instance-dependent component captures the sensitive information about changes in the input while the instance-independent component represents the insensitive information serving solely to minimize the loss on the training dataset. Inspired by the decomposition, we analytically derive a simple extension to current softmax-linear models, which learns to disentangle the two components during training. On several common vision models, the disentangled model out-performs other calibration methods on standard calibration metrics in the face of out-of-distribution (OOD) data and corruption with significantly less complexity. Specifically, we surpass the current state of the art by 30.8% relative improvement on corrupted CIFAR100 in Expected Calibration Error.

While significant advances have been made for single-agent perception, many applications require multiple sensing agents and cross-agent communication due to benefits such as coverage and robustness. It is therefore critical to develop frameworks which support multi-agent collaborative perception in a distributed and bandwidth-efficient manner. In this paper, we address the collaborative perception problem, where one agent is required to perform a perception task and can communicate and share information with other agents on the same task. Specifically, we propose a communication framework by learning both to construct communication groups and decide when to communicate. We demonstrate the generalizability of our framework on two different perception tasks and show that it significantly reduces communication bandwidth while maintaining superior performance.

In this paper, we propose the problem of collaborative perception, where robots can combine their local observations with those of neighboring agents in a learnable way to improve accuracy on a perception task. Unlike existing work in robotics and multi-agent reinforcement learning, we formulate the problem as one where learned information must be shared across a set of agents in a bandwidth-sensitive manner to optimize for scene understanding tasks such as semantic segmentation. Inspired by networking communication protocols, we propose a multi-stage handshake communication mechanism where the neural network can learn to compress relevant information needed for each stage. Specifically, a target agent with degraded sensor data sends a compressed request, the other agents respond with matching scores, and the target agent determines who to connect with (i.e., receive information from). We additionally develop the AirSim-CP dataset and metrics based on the AirSim simulator where a group of aerial robots perceive diverse landscapes, such as roads, grasslands, buildings, etc. We show that for the semantic segmentation task, our handshake communication method significantly improves accuracy by approximately 20% over decentralized baselines, and is comparable to centralized ones using a quarter of the bandwidth.

The fusion of multiple sensor modalities, especially through deep learning architectures, has been an active area of study. However, an under-explored aspect of such work is whether the methods can be robust to degradation across their input modalities, especially when they must generalize to degradation not seen during training. In this work, we propose an uncertainty-aware fusion scheme to effectively fuse inputs that might suffer from a range of known and unknown degradation. Specifically, we analyze a number of uncertainty measures, each of which captures a different aspect of uncertainty, and we propose a novel way to fuse degraded inputs by scaling modality-specific output softmax probabilities. We additionally propose a novel data-dependent spatial temperature scaling method to complement these existing uncertainty measures. Finally, we integrate the uncertainty-scaled output from each modality using a probabilistic noisy-or fusion method. In a photo-realistic simulation environment (AirSim), we show that our method achieves significantly better results on a semantic segmentation task, as compared to state-of-art fusion architectures, on a range of degradation (e.g. fog, snow, frost, and various other types of noise), some of which are unknown during training.

Neural Networks can perform poorly when the training label distribution is heavily imbalanced, as well as when the testing data differs from the training distribution. In order to deal with shift in the testing label distribution, which imbalance causes, we motivate the problem from the perspective of an optimal Bayes classifier and derive a prior rebalancing technique that can be solved through a KL-divergence based optimization. This method allows a flexible post-training hyper-parameter to be efficiently tuned on a validation set and effectively modify the classifier margin to deal with this imbalance. We further combine this method with existing likelihood shift methods, re-interpreting them from the same Bayesian perspective, and demonstrating that our method can deal with both problems in a unified way. The resulting algorithm can be conveniently used on probabilistic classification problems agnostic to underlying architectures. Our results on six different datasets and five different architectures show state of art accuracy, including on large-scale imbalanced datasets such as iNaturalist for classification and Synthia for semantic segmentation. Please see https://github.com/GT-RIPL/UNO-IC.git for implementation.

In image captioning where fluency is an important factor in evaluation, n-gram metrics, sequential models are commonly used; however, sequential models generally result in overgeneralized expressions that lack the details that may be present in an input image. Inspired by the idea of the compositional neural module networks in the visual question answering task, we introduce a hierarchical framework for image captioning that explores both compositionality and sequentiality of natural language. Our algorithm learns to compose a detail-rich sentence by selectively attending to different modules corresponding to unique aspects of each object detected in an input image to include specific descriptions such as counts and color. In a set of experiments on the MSCOCO dataset, the proposed model outperforms a state-of-the art model across multiple evaluation metrics, more importantly, presenting visually interpretable results. Furthermore, the breakdown of subcategories f-scores of the SPICE metric and human evaluation on Amazon Mechanical Turk show that our compositional module networks effectively generate accurate and detailed captions.