论文检索

Deep Learning/Other Representation Learning

Yiming Lei, Zilong Li, Yangyang Li, Junping Zhang, Hongming Shan

Interpreting the decisions of deep learning models has been actively studied since the explosion of deep neural networks. One of the most convincing interpretation approaches is salience-based visual interpretation, such as Grad-CAM, where the generation of attention maps depends merely on categorical labels. Although existing interpretation methods can provide explainable decision clues, they often yield partial correspondence between image and saliency maps due to the limited discriminative information from one-hot labels. This paper develops a Language-Image COnsistency model for explainable image classification, termed LICO, by correlating learnable linguistic prompts with corresponding visual features in a coarse-to-fine manner. Specifically, we first establish a coarse global manifold structure alignment by minimizing the distance between the distributions of image and language features. We then achieve fine-grained saliency maps by applying optimal transport (OT) theory to assign local feature maps with class-specific prompts. Extensive experimental results on eight benchmark datasets demonstrate that the proposed LICO achieves a significant improvement in generating more explainable attention maps in conjunction with existing interpretation methods such as Grad-CAM. Remarkably, LICO improves the classification performance of existing models without introducing any computational overhead during inference.

Deep Learning/Attention Mechanisms

Zhiyu Jin, Xuli Shen, Bin Li, Xiangyang Xue

Diffusion models (DMs) have recently gained attention with state-of-the-art performance in text-to-image synthesis. Abiding by the tradition in deep learning, DMs are trained and evaluated on the images with fixed sizes. However, users are demanding for various images with specific sizes and various aspect ratio. This paper focuses on adapting text-to-image diffusion models to handle such variety while maintaining visual fidelity. First we observe that, during the synthesis, lower resolution images suffer from incomplete object portrayal, while higher resolution images exhibit repetitively disordered presentation. Next, we establish a statistical relationship indicating that attention entropy changes with token quantity, suggesting that models aggregate spatial information in proportion to image resolution. The subsequent interpretation on our observations is that objects are incompletely depicted due to limited spatial information for low resolutions, while repetitively disorganized presentation arises from redundant spatial information for high resolutions. From this perspective, we propose a scaling factor to alleviate the change of attention entropy and mitigate the defective pattern observed. Extensive experimental results validate the efficacy of the proposed scaling factor, enabling models to achieve better visual effects, image quality, and text alignment. Notably, these improvements are achieved without additional training or fine-tuning techniques.

Applications/Computer Vision

Jaemin Cho, Abhay Zala, Mohit Bansal

As large language models have demonstrated impressive performance in many domains, recent works have adopted language models (LMs) as controllers of visual modules for vision-and-language tasks. While existing work focuses on equipping LMs with visual understanding, we propose two novel interpretable/explainable visual programming frameworks for text-to-image (T2I) generation and evaluation. First, we introduce VPGen, an interpretable step-by-step T2I generation framework that decomposes T2I generation into three steps: object/count generation, layout generation, and image generation. We employ an LM to handle the first two steps (object/count generation and layout generation), by finetuning it on text-layout pairs. Our step-by-step T2I generation framework provides stronger spatial control than end-to-end models, the dominant approach for this task. Furthermore, we leverage the world knowledge of pretrained LMs, overcoming the limitation of previous layout-guided T2I works that can only handle predefined object classes. We demonstrate that our VPGen has improved control in counts/spatial relations/scales of objects than state-of-the-art T2I generation models. Second, we introduce VPEval, an interpretable and explainable evaluation framework for T2I generation based on visual programming. Unlike previous T2I evaluations with a single scoring model that is accurate in some skills but unreliable in others, VPEval produces evaluation programs that invoke a set of visual modules that are experts in different skills, and also provides visual+textual explanations of the evaluation results. Our analysis shows that VPEval provides a more human-correlated evaluation for skill-specific and open-ended prompts than widely used single model-based evaluation. We hope that our work encourages future progress on interpretable/explainable generation and evaluation for T2I models.

Reinforcement Learning/Discounted Cost Reward

Yudi Zhang, Yali Du, Biwei Huang, Ziyan Wang, Jun Wang, Meng Fang, Mykola Pechenizkiy

A major challenge in reinforcement learning is to determine which state-action pairs are responsible for future rewards that are delayed. Reward redistribution serves as a solution to re-assign credits for each time step from observed sequences. While the majority of current approaches construct the reward redistribution in an uninterpretable manner, we propose to explicitly model the contributions of state and action from a causal perspective, resulting in an interpretable reward redistribution and preserving policy invariance. In this paper, we start by studying the role of causal generative models in reward redistribution by characterizing the generation of Markovian rewards and trajectory-wise long-term return and further propose a framework, called Generative Return Decomposition (GRD), for policy optimization in delayed reward scenarios. Specifically, GRD first identifies the unobservable Markovian rewards and causal relations in the generative process. Then, GRD makes use of the identified causal generative model to form a compact representation to train policy over the most favorable subspace of the state space of the agent. Theoretically, we show that the unobservable Markovian reward function is identifiable, as well as the underlying causal structure and causal models. Experimental results show that our method outperforms state-of-the-art methods and the provided visualization further demonstrates the interpretability of our method.The project page is located at [https://reedzyd.github.io/GenerativeReturnDecomposition/](https://reedzyd.github.io/GenerativeReturnDecomposition/).

Deep Learning/Other Representation Learning

Antonio Norelli, Marco Fumero, Valentino Maiorca, Luca Moschella, Emanuele Rodolà, Francesco Locatello

CLIP proved that aligning visual and language spaces is key to solving many vision tasks without explicit training, but required to train image and text encoders from scratch on a huge dataset. LiT improved this by only training the text encoder and using a pre-trained vision network. In this paper, we show that a common space can be created without any training at all, using single-domain encoders (trained with or without supervision) and a much smaller amount of image-text pairs. Furthermore, our model has unique properties. Most notably, deploying a new version with updated training samples can be done in a matter of seconds. Additionally, the representations in the common space are easily interpretable as every dimension corresponds to the similarity of the input to a unique entry in the multimodal dataset. Experiments on standard zero-shot visual benchmarks demonstrate the typical transfer ability of image-text models. Overall, our method represents a simple yet surprisingly strong baseline for foundation multi-modal models, raising important questions on their data efficiency and on the role of retrieval in machine learning.

Applications/Computer Vision

Liunian Li, Zi-Yi Dou, Nanyun Peng, Kai-Wei Chang

Recent development in vision-language approaches has instigated a paradigm shift in learning visual recognition models from language supervision. These approaches align objects with language queries (e.g. "a photo of a cat") and thus improve the models' adaptability to novel objects and domains. Recent studies have attempted to query these models with complex language expressions that include specifications of fine-grained details, such as colors, shapes, and relations. However, simply incorporating language descriptions into queries does not guarantee accurate interpretation by the models. In fact, our experiments show that GLIP, a state-of-the-art vision-language model for object detection, often disregards contextual information in the language descriptions and instead relies heavily on detecting objects solely by their names. To tackle the challenge, we propose a new description-conditioned (DesCo) paradigm of learning object recognition models with rich language descriptions consisting of two innovations: 1) we employ a large language model as a commonsense knowledge engine to generate rich language descriptions of objects; 2) we design context-sensitive queries to improve the model's ability in deciphering intricate nuances embedded within descriptions and enforce the model to focus on context rather than object names alone. On two novel object detection benchmarks, LVIS and OminiLabel, under the zero-shot detection setting, our approach achieves 34.8 APr minival (+9.1) and 29.3 AP (+3.6), respectively, surpassing the prior state-of-the-art models, GLIP and FIBER, by a large margin.

Applications/Neuroscience, Cognitive Science

Pierre-Étienne Fiquet, Eero Simoncelli

All organisms make temporal predictions, and their evolutionary fitness level depends on the accuracy of these predictions. In the context of visual perception, the motions of both the observer and objects in the scene structure the dynamics of sensory signals, allowing for partial prediction of future signals based on past ones. Here, we propose a self-supervised representation-learning framework that extracts and exploits the regularities of natural videos to compute accurate predictions. We motivate the polar architecture by appealing to the Fourier shift theorem and its group-theoretic generalization, and we optimize its parameters on next-frame prediction. Through controlled experiments, we demonstrate that this approach can discover the representation of simple transformation groups acting in data. When trained on natural video datasets, our framework achieves better prediction performance than traditional motion compensation and rivals conventional deep networks, while maintaining interpretability and speed. Furthermore, the polar computations can be restructured into components resembling normalized simple and direction-selective complex cell models of primate V1 neurons. Thus, polar prediction offers a principled framework for understanding how the visual system represents sensory inputs in a form that simplifies temporal prediction.

Reinforcement Learning

Fabian Paischer, Thomas Adler, Markus Hofmarcher, Sepp Hochreiter

Reinforcement learning agents deployed in the real world often have to cope with partially observable environments. Therefore, most agents employ memory mechanisms to approximate the state of the environment. Recently, there have been impressive success stories in mastering partially observable environments, mostly in the realm of computer games like Dota 2, StarCraft II, or MineCraft. However, existing methods lack interpretability in the sense that it is not comprehensible for humans what the agent stores in its memory.In this regard, we propose a novel memory mechanism that represents past events in human language.Our method uses CLIP to associate visual inputs with language tokens. Then we feed these tokens to a pretrained language model that serves the agent as memory and provides it with a coherent and human-readable representation of the past.We train our memory mechanism on a set of partially observable environments and find that it excels on tasks that require a memory component, while mostly attaining performance on-par with strong baselines on tasks that do not. On a challenging continuous recognition task, where memorizing the past is crucial, our memory mechanism converges two orders of magnitude faster than prior methods.Since our memory mechanism is human-readable, we can peek at an agent's memory and check whether crucial pieces of information have been stored.This significantly enhances troubleshooting and paves the way toward more interpretable agents.

Taotao Jing, Haifeng Xia, Renran Tian, Haoran Ding, Xiao Luo, Joshua Domeyer, Rini Sherony, Zhengming Ding

Autonomous driving has attracted interest for interpretable action decision models that mimic human cognition. Existing interpretable autonomous driving models explore static human explanations, which ignore the implicit visual semantics that are not explicitly annotated or even consistent across annotators. In this paper, we propose a novel Interpretable Action decision making (InAction) model to provide an enriched explanation from both explicit human annotation and implicit visual semantics. First, a proposed visual-semantic module captures the region-based action-inducing components from the visual inputs, which learns the implicit visual semantics to provide a human-understandable explanation in action decision making. Second, an explicit reasoning module is developed by incorporating global visual features and action-inducing visual semantics, which aims to jointly align the human-annotated explanation and action decision making. Experimental results on two autonomous driving benchmarks demonstrate the effectiveness of our InAction model for explaining both implicitly and explicitly by comparing it to existing interpretable autonomous driving models.

Mingyu Yang, Yu Chen, Hun-Seok Kim

In recent years, deep learning-based approaches for visual-inertial odometry (VIO) have shown remarkable performance outperforming traditional geometric methods. Yet, all existing methods use both the visual and inertial measurements for every pose estimation incurring potential computational redundancy. While visual data processing is much more expensive than that for the inertial measurement unit (IMU), it may not always contribute to improving the pose estimation accuracy. In this paper, we propose an adaptive deep-learning based VIO method that reduces computational redundancy by opportunistically disabling the visual modality. Specifically, we train a policy network that learns to deactivate the visual feature extractor on the fly based on the current motion state and IMU readings. A Gumbel-Softmax trick is adopted to train the policy network to make the decision process differentiable for end-to-end system training. The learned strategy is interpretable, and it shows scenario-dependent decision patterns for adaptive complexity reduction. Experiment results show that our method achieves a similar or even better performance than the full-modality baseline with up to 78.8% computational complexity reduction for KITTI dataset evaluation. The code is available at https://github.com/mingyuyng/Visual-Selective-VIO.

Ruocheng Wang, Yunzhi Zhang, Jiayuan Mao, Chin-Yi Cheng, Jiajun Wu

We study the problem of translating an image-based, step-by-step assembly manual created by human designers into machine-interpretable instructions. We formulate this problem as a sequential prediction task: at each step, our model reads the manual, locates the components to be added to the current shape, and infers their 3D poses. This task poses the challenge of establishing a 2D-3D correspondence between the manual image and the real 3D object, and 3D pose estimation for unseen 3D objects, since a new component to be added in a step can be an object built from previous steps. To address these two challenges, we present a novel learning-based framework, the Manual-to-Executable-Plan Network (MEPNet), which reconstructs the assembly steps from a sequence of manual images. The key idea is to integrate neural 2D keypoint detection modules and 2D-3D projection algorithms for high-precision prediction and strong generalization to unseen components. The MEPNet outperforms existing methods on three newly collected LEGO manual datasets and a Minecraft house dataset.

Karren Yang, Michael Firman, Eric Brachmann, Clément Godard

In this work, we show how to estimate a device’s position and orientation indoors by echolocation, i.e., by interpreting the echoes of an audio signal that the device itself emits. Established visual localization methods rely on the device’s camera and yield excellent accuracy if unique visual features are in view and depicted clearly. We argue that audio sensing can offer complementary information to vision for device localization, since audio is invariant to adverse visual conditions and can reveal scene information beyond a camera’s field of view. We first propose a strategy for learning an audio representation that captures the scene geometry around a device using supervision transfer from vision. Subsequently, we leverage this audio representation to complement vision in three device localization tasks: relative pose estimation, place recognition, and absolute pose regression. Our proposed methods outperform state-of-the-art vision models on new audio-visual benchmarks for the Replica and Matterport3D datasets.

Zhengyuan Yang, Zhe Gan, Jianfeng Wang, Xiaowei Hu, Faisal Ahmed, Zicheng Liu, Yumao Lu, Lijuan Wang

We propose UniTAB that Unifies Text And Box outputs for grounded vision-language (VL) modeling. Grounded VL tasks such as grounded captioning require the model to generate a text description and align predicted words with object regions. To achieve this, models must generate desired text and box outputs together, and meanwhile indicate the alignments between words and boxes. In contrast to existing solutions that use multiple separate modules for different outputs, UniTAB represents both text and box outputs with a shared token sequence, and introduces a special token to naturally indicate word-box alignments in the sequence. UniTAB thus could provide a more comprehensive and interpretable image description, by freely grounding generated words to object regions. On grounded captioning, UniTAB presents a simpler solution with a single output head, and significantly outperforms state of the art in both grounding and captioning evaluations. On general VL tasks that have different desired output formats (i.e., text, box, or their combination), UniTAB with a single network achieves better or comparable performance than task-specific state of the art. Experiments cover 7 VL benchmarks, including grounded captioning, visual grounding, image captioning, and visual question answering. Furthermore, UniTAB’s unified multi-task network and the task-agnostic output sequence design make the model parameter efficient and generalizable to new tasks.

Jiafei Duan, Samson Yu, Soujanya Poria, Bihan Wen, Cheston Tan

Accurate prediction of physical interaction outcomes is a crucial component of human intelligence and is important for safe and efficient deployments of robots in the real world. While there are existing vision-based intuitive physics models that learn to predict physical interaction outcomes, they mostly focus on generating short sequences of future frames based on physical properties (e.g. mass, friction and velocity) extracted from visual inputs or a latent space. However, there is a lack of intuitive physics models that are tested on long physical interaction sequences with multiple interactions among different objects. We hypothesize that selective temporal attention during approximate mental simulations helps humans in physical interaction outcome prediction. With these motivations, we propose a novel scheme: \textbf{P}hysical \textbf{I}nteraction \textbf{P}rediction via Mental Simulation with Span Selection (PIP). It utilizes a deep generative model to model approximate mental simulations by generating future frames of physical interactions before employing selective temporal attention in the form of span selection for predicting physical interaction outcomes. To the best of our knowledge, attention has not been used with deep learning to tackle intuitive physics. For model evaluation, we further propose the large-scale SPACE+ dataset of synthetic videos with long sequences of three prime physical interactions in a 3D environment. Our experiments show that PIP outperforms human, baseline, and related intuitive physics models that utilize mental simulation. Furthermore, PIP’s span selection module effectively identifies the frames indicating key physical interactions among objects, allowing for added interpretability, and does not require labor-intensive frame annotations. PIP is available on https://sites.google.com/view/piphysics.

Honglu Zhou, Asim Kadav, Aviv Shamsian, Shijie Geng, Farley Lai, Long Zhao, Ting Liu, Mubbasir Kapadia, Hans Peter Graf

Group Activity Recognition detects the activity collectively performed by a group of actors, which requires compositional reasoning of actors and objects. We approach the task by modeling the video as tokens that represent the multi-scale semantic concepts in the video. We propose COMPOSER, a Multiscale Transformer based architecture that performs attention-based reasoning over tokens at each scale and learns group activity compositionally. In addition, prior works suffer from scene biases with privacy and ethical concerns. We only use the keypoint modality which reduces scene biases and prevents acquiring detailed visual data that may contain private or biased information of users. We improve the multiscale representations in COMPOSER by clustering the intermediate scale representations, while maintaining consistent cluster assignments between scales. Finally, we use techniques such as auxiliary prediction and data augmentations tailored to the keypoint signals to aid model training. We demonstrate the model’s strength and interpretability on two widely-used datasets (Volleyball and Collective Activity). COMPOSER achieves up to +5.4% improvement with just the keypoint modality.

Xinhao Li, Jingjing Li, Zhekai Du, Lei Zhu, Wen Li

Open-set Domain Adaptation (OSDA) aims to recognize classes in the target domain that are seen in the source domain while rejecting other unseen target-exclusive classes into an unknown class, which ignores the diversity of the latter and is therefore incapable of their interpretation. The recently-proposed Semantic Recovery OSDA (SR-OSDA) brings in semantic attributes and attacks the challenge via partial alignment and visual-semantic projection, marking the first step towards interpretable OSDA. Following that line, in this work, we propose a representation learning framework termed Angular Margin Separation (AMS) that unveils the power of discriminative and robust representation for both open-set domain adaptation and cross-domain semantic recovery. Our core idea is to exploit an additive angular margin with regularization for both robust feature fine-tuning and discriminative joint feature alignment, which turns out advantageous to learning an accurate and less biased visual-semantic projection. Further, we propose a post-training re-projection that boosts the performance of seen classes interpretation without deterioration on unseen classes. Verified by extensive experiments, AMS achieves a notable improvement over the existing SR-OSDA baseline, with an average 7.6% increment in semantic recovery accuracy of unseen classes in multiple transfer tasks. Our code is available at https://github.com/LeoXinhaoLee/AMS.

Cheng Da, Peng Wang, Cong Yao

A novel scene text recognizer based on Vision-Language Transformer (VLT) is presented. Inspired by Levenshtein Transformer in the area of NLP, the proposed method (named Levenshtein OCR, and LevOCR for short) explores an alternative way for automatically transcribing textual content from cropped natural images. Specifically, we cast the problem of scene text recognition as an iterative sequence refinement process. The initial prediction sequence produced by a pure vision model is encoded and fed into a cross-modal transformer to interact and fuse with the visual features, to progressively approximate the ground truth. The refinement process is accomplished via two basic characterlevel operations: deletion and insertion, which are learned with imitation learning and allow for parallel decoding, dynamic length change and good interpretability. The quantitative experiments clearly demonstrate that LevOCR achieves state-of-the-art performances on standard benchmarks and the qualitative analyses verify the effectiveness and advantage of the proposed LevOCR algorithm.

Moinak Bhattacharya, Shubham Jain, Prateek Prasanna

In this work, we present RadioTransformer, a novel visual attention-driven transformer framework, that leverages radiologists’ gaze patterns and models their visuo-cognitive behavior for disease diagnosis on chest radiographs. Domain experts, such as radiologists, rely on visual information for medical image interpretation. On the other hand, deep neural networks have demonstrated significant promise in similar tasks even where visual interpretation is challenging. Eye-gaze tracking has been used to capture the viewing behavior of domain experts, lending insights into the complexity of visual search. However, deep learning frameworks, even those that rely on attention mechanisms, do not leverage this rich domain information. RadioTransformer fills this critical gap by learning from radiologists’ visual search patterns, encoded as ‘human visual attention regions’ in a cascaded global-focal transformer framework. The overall ‘global’ image characteristics and the more detailed ‘local’ features are captured by the proposed global and focal modules, respectively. We experimentally validate the efficacy of our student-teacher approach for 8 datasets involving different disease classification tasks where eye-gaze data is not available during the inference phase. Code: https://github.com/bmi-imaginelab/radiotransformer.

Jiezhang Cao, Jingyun Liang, Kai Zhang, Wenguan Wang, Qin Wang, Yulun Zhang, Hao Tang, Luc Van Gool

In this paper, we study a practical space-time video super-resolution (STVSR) problem which aims at generating a high-framerate high-resolution sharp video from a low-framerate low-resolution blurry video. Such problem often occurs when recording a fast dynamic event with a low-framerate and low-resolution camera, and the captured video would suffer from three typical issues: i) motion blur occurs due to object/camera motions during exposure time; ii) motion aliasing is unavoidable when the event temporal frequency exceeds the Nyquist limit of temporal sampling; iii) high-frequency details are lost because of the low spatial sampling rate. These issues can be alleviated by a cascade of three separate sub-tasks, including video deblurring, frame interpolation, and super-resolution, which, however, would fail to capture the spatial and temporal correlations among video sequences. To address this, we propose an interpretable STVSR framework by leveraging both model-based and learning-based methods. Specifically, we formulate STVSR as a joint video deblurring, frame interpolation, and super-resolution problem, and solve it as two sub-problems in an alternate way. For the first sub-problem, we derive an interpretable analytical solution and use it as a Fourier data transform layer. Then, we propose a recurrent video enhancement layer for the second sub-problem to further recover high-frequency details. Extensive experiments demonstrate the superiority of our method in terms of quantitative metrics and visual quality.

Nan Liu, Shuang Li, Yilun Du, Antonio Torralba, Joshua B. Tenenbaum

Large text-guided diffusion models, such as DALLE-2, are able to generate stunning photorealistic images given natural language descriptions. While such models are highly flexible, they struggle to understand the composition of certain concepts, such as confusing the attributes of different objects or relations between objects. In this paper, we propose an alternative structured approach for compositional generation using diffusion models. An image is generated by composing a set of diffusion models, with each of them modeling a certain component of the image. To do this, we interpret diffusion models as energy-based models in which the data distributions defined by the energy functions may be explicitly combined. The proposed method can generate scenes at test time that are substantially more complex than those seen in training, composing sentence descriptions, object relations, human facial attributes, and even generalizing to new combinations that are rarely seen in the real world. We further illustrate how our approach may be used to compose pre-trained text-guided diffusion models and generate photorealistic images containing all the details described in the input descriptions, including the binding of certain object attributes that have been shown difficult for DALLE-2. These results point to the effectiveness of the proposed method in promoting structured generalization for visual generation.