Abstract
We present a unified probabilistic model that learns a representative set of discrete vehicle actions and predicts the probability of each action given a particular scenario. Our model also enables us to estimate the distribution over continuous trajectories conditioned on a scenario, representing what each discrete action would look like if executed in that scenario. While our primary objective is to learn representative action sets, these capabilities combine to produce accurate multimodal trajectory predictions as a byproduct. Although our learned action representations closely resemble semantically meaningful categories (e.g., "go straight", "turn left", etc.), our method is entirely self-supervised and does not utilize any manually generated labels or categories. Our method builds upon recent advances in variational inference and deep unsupervised clustering, resulting in full distribution estimates based on deterministic model evaluations.
Establishing a stronger baseline for lightweight contrastive models
Authors: Wenye Lin, Yifeng Ding, Zhixiong Cao, Hai-tao Zheng
Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Abstract
Recent research has reported a performance degradation in self-supervised contrastive learning for specially designed efficient networks, such as MobileNet and EfficientNet. A common practice to address this problem is to introduce a pretrained contrastive teacher model and train the lightweight networks with distillation signals generated by the teacher. However, it is time and resource consuming to pretrain a teacher model when it is not available. In this work, we aim to establish a stronger baseline for lightweight contrastive models without using a pretrained teacher model. Specifically, we show that the optimal recipe for efficient models is different from that of larger models, and using the same training settings as ResNet50, as previous research does, is inappropriate. Additionally, we observe a common issu e in contrastive learning where either the positive or negative views can be noisy, and propose a smoothed version of InfoNCE loss to alleviate this problem. As a result, we successfully improve the linear evaluation results from 36.3\% to 62.3\% for MobileNet-V3-Large and from 42.2\% to 65.8\% for EfficientNet-B0 on ImageNet, closing the accuracy gap to ResNet50 with $5\times$ fewer parameters. We hope our research will facilitate the usage of lightweight contrastive models.
Mitigating Negative Style Transfer in Hybrid Dialogue System
Abstract
As the functionality of dialogue systems evolves, hybrid dialogue systems that accomplish user-specific goals and participate in open-topic chitchat with users are attracting growing attention. Existing research learns both tasks concurrently utilizing a multi-task fusion technique but ignores the negative transfer phenomenon induced by the unique textual style differences. Therefore, contrastive learning based on the latent variable model is used to decouple the various textual genres in the latent space. We devise supervised and self-supervised positive and negative sample constructions for diverse datasets. In addition, to capitalize on the style information contained in the decoupled latent variables, we employ a style prefix that incorporates latent variables further to control the generation of responses with varying styles. We performed extensive experiments on three dialogue datasets, including a hybrid dialogue dataset and two task-oriented dialogue datasets. The experimental results demonstrate that our method can mitigate the negative style transfer issue and achieves state-of-the-art performance on multiple dialogue datasets.
MAELi -- Masked Autoencoder for Large-Scale LiDAR Point Clouds
Authors: Georg Krispel, David Schinagl, Christian Fruhwirth-Reisinger, Horst Possegger, Horst Bischof
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
We show how the inherent, but often neglected, properties of large-scale LiDAR point clouds can be exploited for effective self-supervised representation learning. To this end, we design a highly data-efficient feature pre-training backbone that significantly reduces the amount of tedious 3D annotations to train state-of-the-art object detectors. In particular, we propose a Masked AutoEncoder (MAELi) that intuitively utilizes the sparsity of the LiDAR point clouds in both, the encoder and the decoder, during reconstruction. This results in more expressive and useful features, directly applicable to downstream perception tasks, such as 3D object detection for autonomous driving. In a novel reconstruction scheme, MAELi distinguishes between free and occluded space and leverages a new masking strategy which targets the LiDAR's inherent spherical projection. To demonstrate the potential of MAELi, we pre-train one of the most widespread 3D backbones, in an end-to-end fashion and show the merit of our fully unsupervised pre-trained features on several 3D object detection architectures. Given only a tiny fraction of labeled frames to fine-tune such detectors, we achieve significant performance improvements. For example, with only $\sim800$ labeled frames, MAELi features improve a SECOND model by +10.09APH/LEVEL 2 on Waymo Vehicles.
Image Compression with Product Quantized Masked Image Modeling
Authors: Alaaeldin El-Nouby, Matthew J. Muckley, Karen Ullrich, Ivan Laptev, Jakob Verbeek, Hervé Jégou
Subjects: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
Abstract
Recent neural compression methods have been based on the popular hyperprior framework. It relies on Scalar Quantization and offers a very strong compression performance. This contrasts from recent advances in image generation and representation learning, where Vector Quantization is more commonly employed. In this work, we attempt to bring these lines of research closer by revisiting vector quantization for image compression. We build upon the VQ-VAE framework and introduce several modifications. First, we replace the vanilla vector quantizer by a product quantizer. This intermediate solution between vector and scalar quantization allows for a much wider set of rate-distortion points: It implicitly defines high-quality quantizers that would otherwise require intractably large codebooks. Second, inspired by the success of Masked Image Modeling (MIM) in the context of self-supervised learning and generative image models, we propose a novel conditional entropy model which improves entropy coding by modelling the co-dependencies of the quantized latent codes. The resulting PQ-MIM model is surprisingly effective: its compression performance on par with recent hyperprior methods. It also outperforms HiFiC in terms of FID and KID metrics when optimized with perceptual losses (e.g. adversarial). Finally, since PQ-MIM is compatible with image generation frameworks, we show qualitatively that it can operate under a hybrid mode between compression and generation, with no further training or finetuning. As a result, we explore the extreme compression regime where an image is compressed into 200 bytes, i.e., less than a tweet.
BKinD-3D: Self-Supervised 3D Keypoint Discovery from Multi-View Videos
Authors: Jennifer J. Sun, Pierre Karashchuk, Amil Dravid, Serim Ryou, Sonia Fereidooni, John Tuthill, Aggelos Katsaggelos, Bingni W. Brunton, Georgia Gkioxari, Ann Kennedy, Yisong Yue, Pietro Perona
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
Abstract
Quantifying motion in 3D is important for studying the behavior of humans and other animals, but manual pose annotations are expensive and time-consuming to obtain. Self-supervised keypoint discovery is a promising strategy for estimating 3D poses without annotations. However, current keypoint discovery approaches commonly process single 2D views and do not operate in the 3D space. We propose a new method to perform self-supervised keypoint discovery in 3D from multi-view videos of behaving agents, without any keypoint or bounding box supervision in 2D or 3D. Our method uses an encoder-decoder architecture with a 3D volumetric heatmap, trained to reconstruct spatiotemporal differences across multiple views, in addition to joint length constraints on a learned 3D skeleton of the subject. In this way, we discover keypoints without requiring manual supervision in videos of humans and rats, demonstrating the potential of 3D keypoint discovery for studying behavior.
Keyword: vision transformer
There is no result
Keyword: multimodal
Learning and Predicting Multimodal Vehicle Action Distributions in a Unified Probabilistic Model Without Labels
Authors: Charles Richter, Patrick R. Barragán, Sertac Karaman
Abstract
We present a unified probabilistic model that learns a representative set of discrete vehicle actions and predicts the probability of each action given a particular scenario. Our model also enables us to estimate the distribution over continuous trajectories conditioned on a scenario, representing what each discrete action would look like if executed in that scenario. While our primary objective is to learn representative action sets, these capabilities combine to produce accurate multimodal trajectory predictions as a byproduct. Although our learned action representations closely resemble semantically meaningful categories (e.g., "go straight", "turn left", etc.), our method is entirely self-supervised and does not utilize any manually generated labels or categories. Our method builds upon recent advances in variational inference and deep unsupervised clustering, resulting in full distribution estimates based on deterministic model evaluations.
Design-time Fashion Popularity Forecasting in VR Environments
Abstract
Being able to forecast the popularity of new garment designs is very important in an industry as fast paced as fashion, both in terms of profitability and reducing the problem of unsold inventory. Here, we attempt to address this task in order to provide informative forecasts to fashion designers within a virtual reality designer application that will allow them to fine tune their creations based on current consumer preferences within an interactive and immersive environment. To achieve this we have to deal with the following central challenges: (1) the proposed method should not hinder the creative process and thus it has to rely only on the garment's visual characteristics, (2) the new garment lacks historical data from which to extrapolate their future popularity and (3) fashion trends in general are highly dynamical. To this end, we develop a computer vision pipeline fine tuned on fashion imagery in order to extract relevant visual features along with the category and attributes of the garment. We propose a hierarchical label sharing (HLS) pipeline for automatically capturing hierarchical relations among fashion categories and attributes. Moreover, we propose MuQAR, a Multimodal Quasi-AutoRegressive neural network that forecasts the popularity of new garments by combining their visual features and categorical features while an autoregressive neural network is modelling the popularity time series of the garment's category and attributes. Both the proposed HLS and MuQAR prove capable of surpassing the current state-of-the-art in key benchmark datasets, DeepFashion for image classification and VISUELLE for new garment sales forecasting.
Modeling Multimodal Aleatoric Uncertainty in Segmentation with Mixture of Stochastic Expert
Authors: Zhitong Gao, Yucong Chen, Chuyu Zhang, Xuming He
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Equipping predicted segmentation with calibrated uncertainty is essential for safety-critical applications. In this work, we focus on capturing the data-inherent uncertainty (aka aleatoric uncertainty) in segmentation, typically when ambiguities exist in input images. Due to the high-dimensional output space and potential multiple modes in segmenting ambiguous images, it remains challenging to predict well-calibrated uncertainty for segmentation. To tackle this problem, we propose a novel mixture of stochastic experts (MoSE) model, where each expert network estimates a distinct mode of the aleatoric uncertainty and a gating network predicts the probabilities of an input image being segmented in those modes. This yields an efficient two-level uncertainty representation. To learn the model, we develop a Wasserstein-like loss that directly minimizes the distribution distance between the MoSE and ground truth annotations. The loss can easily integrate traditional segmentation quality measures and be efficiently optimized via constraint relaxation. We validate our method on the LIDC-IDRI dataset and a modified multimodal Cityscapes dataset. Results demonstrate that our method achieves the state-of-the-art or competitive performance on all metrics.
Keyword: CLIP
LidarCLIP or: How I Learned to Talk to Point Clouds
Authors: Georg Hess, Adam Tonderski, Christoffer Petersson, Lennart Svensson, Kalle Åström
Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Abstract
Research connecting text and images has recently seen several breakthroughs, with models like CLIP, DALL-E 2, and Stable Diffusion. However, the connection between text and other visual modalities, such as lidar data, has received less attention, prohibited by the lack of text-lidar datasets. In this work, we propose LidarCLIP, a mapping from automotive point clouds to a pre-existing CLIP embedding space. Using image-lidar pairs, we supervise a point cloud encoder with the image CLIP embeddings, effectively relating text and lidar data with the image domain as an intermediary. We show the effectiveness of LidarCLIP by demonstrating that lidar-based retrieval is generally on par with image-based retrieval, but with complementary strengths and weaknesses. By combining image and lidar features, we improve upon both single-modality methods and enable a targeted search for challenging detection scenarios under adverse sensor conditions. We also use LidarCLIP as a tool to investigate fundamental lidar capabilities through natural language. Finally, we leverage our compatibility with CLIP to explore a range of applications, such as point cloud captioning and lidar-to-image generation, without any additional training. We hope LidarCLIP can inspire future work to dive deeper into connections between text and point cloud understanding. Code and trained models available at https://github.com/atonderski/lidarclip.
Localizing Objects in 3D from Egocentric Videos with Visual Queries
Authors: Jinjie Mai, Abdullah Hamdi, Silvio Giancola, Chen Zhao, Bernard Ghanem
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
With the recent advances in video and 3D understanding, novel 4D spatio-temporal challenges fusing both concepts have emerged. Towards this direction, the Ego4D Episodic Memory Benchmark proposed a task for Visual Queries with 3D Localization (VQ3D). Given an egocentric video clip and an image crop depicting a query object, the goal is to localize the 3D position of the center of that query object with respect to the camera pose of a query frame. Current methods tackle the problem of VQ3D by lifting the 2D localization results of the sister task Visual Queries with 2D Localization (VQ2D) into a 3D reconstruction. Yet, we point out that the low number of Queries with Poses (QwP) from previous VQ3D methods severally hinders their overall success rate and highlights the need for further effort in 3D modeling to tackle the VQ3D task. In this work, we formalize a pipeline that better entangles 3D multiview geometry with 2D object retrieval from egocentric videos. We estimate more robust camera poses, leading to more successful object queries and substantially improved VQ3D performance. In practice, our method reaches a top-1 overall success rate of 86.36% on the Ego4D Episodic Memory Benchmark VQ3D, a 10x improvement over the previous state-of-the-art. In addition, we provide a complete empirical study highlighting the remaining challenges in VQ3D.
Understanding Zero-Shot Adversarial Robustness for Large-Scale Models
Authors: Chengzhi Mao, Scott Geng, Junfeng Yang, Xin Wang, Carl Vondrick
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Pretrained large-scale vision-language models like CLIP have exhibited strong generalization over unseen tasks. Yet imperceptible adversarial perturbations can significantly reduce CLIP's performance on new tasks. In this work, we identify and explore the problem of \emph{adapting large-scale models for zero-shot adversarial robustness}. We first identify two key factors during model adaption -- training losses and adaptation methods -- that affect the model's zero-shot adversarial robustness. We then propose a text-guided contrastive adversarial training loss, which aligns the text embeddings and the adversarial visual features with contrastive learning on a small set of training data. We apply this training loss to two adaption methods, model finetuning and visual prompt tuning. We find that visual prompt tuning is more effective in the absence of texts, while finetuning wins in the existence of text guidance. Overall, our approach significantly improves the zero-shot adversarial robustness over CLIP, seeing an average improvement of over 31 points over ImageNet and 15 zero-shot datasets. We hope this work can shed light on understanding the zero-shot adversarial robustness of large-scale models.
Abstract
Deep neural networks have been successfully adopted to diverse domains including pathology classification based on medical images. However, large-scale and high-quality data to train powerful neural networks are rare in the medical domain as the labeling must be done by qualified experts. Researchers recently tackled this problem with some success by taking advantage of models pre-trained on large-scale general domain data. Specifically, researchers took contrastive image-text encoders (e.g., CLIP) and fine-tuned it with chest X-ray images and paired reports to perform zero-shot pathology classification, thus completely removing the need for pathology-annotated images to train a classification model. Existing studies, however, fine-tuned the pre-trained model with the same contrastive learning objective, and failed to exploit the multi-labeled nature of medical image-report pairs. In this paper, we propose a new fine-tuning strategy based on sentence sampling and positive-pair loss relaxation for improving the downstream zero-shot pathology classification performance, which can be applied to any pre-trained contrastive image-text encoders. Our method consistently showed dramatically improved zero-shot pathology classification performance on four different chest X-ray datasets and 3 different pre-trained models (5.77% average AUROC increase). In particular, fine-tuning CLIP with our method showed much comparable or marginally outperformed to board-certified radiologists (0.619 vs 0.625 in F1 score and 0.530 vs 0.544 in MCC) in zero-shot classification of five prominent diseases from the CheXpert dataset.
CLIPSep: Learning Text-queried Sound Separation with Noisy Unlabeled Videos
Abstract
Recent years have seen progress beyond domain-specific sound separation for speech or music towards universal sound separation for arbitrary sounds. Prior work on universal sound separation has investigated separating a target sound out of an audio mixture given a text query. Such text-queried sound separation systems provide a natural and scalable interface for specifying arbitrary target sounds. However, supervised text-queried sound separation systems require costly labeled audio-text pairs for training. Moreover, the audio provided in existing datasets is often recorded in a controlled environment, causing a considerable generalization gap to noisy audio in the wild. In this work, we aim to approach text-queried universal sound separation by using only unlabeled data. We propose to leverage the visual modality as a bridge to learn the desired audio-textual correspondence. The proposed CLIPSep model first encodes the input query into a query vector using the contrastive language-image pretraining (CLIP) model, and the query vector is then used to condition an audio separation model to separate out the target sound. While the model is trained on image-audio pairs extracted from unlabeled videos, at test time we can instead query the model with text inputs in a zero-shot setting, thanks to the joint language-image embedding learned by the CLIP model. Further, videos in the wild often contain off-screen sounds and background noise that may hinder the model from learning the desired audio-textual correspondence. To address this problem, we further propose an approach called noise invariant training for training a query-based sound separation model on noisy data. Experimental results show that the proposed models successfully learn text-queried universal sound separation using only noisy unlabeled videos, even achieving competitive performance against a supervised model in some settings.
Abstract
Large-scale cross-modal pre-training paradigms have recently shown ubiquitous success on a wide range of downstream tasks, e.g., zero-shot classification, retrieval and image captioning. However, their successes highly rely on the scale and quality of web-crawled data that naturally contain incomplete and noisy information (e.g., wrong or irrelevant content). Existing works either design manual rules to clean data or generate pseudo-targets as auxiliary signals for reducing noise impact, which do not explicitly tackle both the incorrect and incomplete challenges simultaneously. In this paper, to automatically mitigate the impact of noise by solely mining over existing data, we propose a principled Noise-robust Language-Image Pre-training framework (NLIP) to stabilize pre-training via two schemes: noise-harmonization and noise-completion. First, in noise-harmonization scheme, NLIP estimates the noise probability of each pair according to the memorization effect of cross-modal transformers, then adopts noise-adaptive regularization to harmonize the cross-modal alignments with varying degrees. Second, in noise-completion scheme, to enrich the missing object information of text, NLIP injects a concept-conditioned cross-modal decoder to obtain semantic-consistent synthetic captions to complete noisy ones, which uses the retrieved visual concepts (i.e., objects' names) for the corresponding image to guide captioning generation. By collaboratively optimizing noise-harmonization and noise-completion schemes, our NLIP can alleviate the common noise effects during image-text pre-training in a more efficient way. Extensive experiments show the significant performance improvements of our NLIP using only 26M data over existing pre-trained models (e.g., CLIP, FILIP and BLIP) on 12 zero-shot classification datasets, MSCOCO image captioning and zero-shot image-text retrieval tasks.
Reproducible scaling laws for contrastive language-image learning
Authors: Mehdi Cherti, Romain Beaumont, Ross Wightman, Mitchell Wortsman, Gabriel Ilharco, Cade Gordon, Christoph Schuhmann, Ludwig Schmidt, Jenia Jitsev
Abstract
Scaling up neural networks has led to remarkable performance across a wide range of tasks. Moreover, performance often follows reliable scaling laws as a function of training set size, model size, and compute, which offers valuable guidance as large-scale experiments are becoming increasingly expensive. However, previous work on scaling laws has primarily used private data \& models or focused on uni-modal language or vision learning. To address these limitations, we investigate scaling laws for contrastive language-image pre-training (CLIP) with the public LAION dataset and the open-source OpenCLIP repository. Our large-scale experiments involve models trained on up to two billion image-text pairs and identify power law scaling for multiple downstream tasks including zero-shot classification, retrieval, linear probing, and end-to-end fine-tuning. We find that the training distribution plays a key role in scaling laws as the OpenAI and OpenCLIP models exhibit different scaling behavior despite identical model architectures and similar training recipes. We open-source our evaluation workflow and all models, including the largest public CLIP models, to ensure reproducibility and make scaling laws research more accessible. Source code and instructions to reproduce this study will be available at https://github.com/LAION-AI/scaling-laws-openclip
Keyword: metric learning
There is no result
Keyword: image retrieval
There is no result
Keyword: self-supervised
Learning and Predicting Multimodal Vehicle Action Distributions in a Unified Probabilistic Model Without Labels
Establishing a stronger baseline for lightweight contrastive models
Mitigating Negative Style Transfer in Hybrid Dialogue System
MAELi -- Masked Autoencoder for Large-Scale LiDAR Point Clouds
Image Compression with Product Quantized Masked Image Modeling
BKinD-3D: Self-Supervised 3D Keypoint Discovery from Multi-View Videos
Keyword: vision transformer
There is no result
Keyword: multimodal
Learning and Predicting Multimodal Vehicle Action Distributions in a Unified Probabilistic Model Without Labels
Design-time Fashion Popularity Forecasting in VR Environments
Modeling Multimodal Aleatoric Uncertainty in Segmentation with Mixture of Stochastic Expert
Keyword: CLIP
LidarCLIP or: How I Learned to Talk to Point Clouds
Localizing Objects in 3D from Egocentric Videos with Visual Queries
Understanding Zero-Shot Adversarial Robustness for Large-Scale Models
Significantly improving zero-shot X-ray pathology classification via fine-tuning pre-trained image-text encoders
CLIPSep: Learning Text-queried Sound Separation with Noisy Unlabeled Videos
NLIP: Noise-robust Language-Image Pre-training
Reproducible scaling laws for contrastive language-image learning
Keyword: DALLE
There is no result