This paper aims to interpret how deepfake detection models learn artifact features of images when just supervised by binary labels. To this end, three hypotheses from the perspective of image matching are proposed as follows. 1. Deepfake detection models indicate real/fake images based on visual concepts that are neither source-relevant nor target-relevant, that is, considering such visual concepts as artifact-relevant. 2. Besides the supervision of binary labels, deepfake detection models implicitly learn artifact-relevant visual concepts through the FST-Matching (i.e. the matching fake, source, target images) in the training set. 3. Implicitly learned artifact visual concepts through the FST-Matching in the raw training set are vulnerable to video compression. In experiments, the above hypotheses are verified among various DNNs. Furthermore, based on this understanding, we propose the FST-Matching Deepfake Detection Model to boost the performance of forgery detection on compressed videos. Experiment results show that our method achieves great performance, especially on highly-compressed (e.g. c40) videos.
论文检索
输入标题、作者或关键词，从 7,537 篇学术成果中精准定位
3D face reconstruction and facial action unit (AU) detection have emerged as interesting and challenging tasks in recent years, but are rarely performed in tandem. Image-based 3D face reconstruction, which can represent a dense space of facial motions, is typically accomplished by estimating identity, expression, texture, head pose, and illumination separately via pre-constructed 3D morphable models (3DMMs). Recent 3D reconstruction models can recover high-quality geometric facial details like wrinkles and pores, but are still limited in their ability to recover 3D subtle motions caused by the activation of AUs. We present a multi-stage learning framework that recovers AU-interpretable 3D facial details by learning personalized AU-specific blendshapes from images. Our model explicitly learns 3D expression basis by using AU labels and generic AU relationship prior and then constrains the basis coefficients such that they are semantically mapped to each AU. Our AU-aware 3D reconstruction model generates accurate 3D expressions composed by semantically meaningful AU motion components. Furthermore, the output of the model can be directly applied to generate 3D AU occurrence predictions, which have not been fully explored by prior 3D reconstruction models. We demonstrate the effectiveness of our approach via qualitative and quantitative evaluations.
Explaining deep convolutional neural networks has been recently drawing increasing attention since it helps to understand the networks’ internal operations and why they make certain decisions. Saliency maps, which emphasize salient regions largely connected to the network’s decision-making, are one of the most common ways for visualizing and analyzing deep networks in the computer vision community. However, saliency maps generated by existing methods cannot represent authentic information in images due to the unproven proposals about the weights of activation maps which lack solid theoretical foundation and fail to consider the relations between each pixels. In this paper, we develop a novel post-hoc visual explanation method called Shap-CAM based on class activation mapping. Unlike previous gradient-based approaches, Shap-CAM gets rid of the dependence on gradients by obtaining the importance of each pixels through Shapley value. We demonstrate that Shap-CAM achieves better visual performance and fairness for interpreting the decision making process. Our approach outperforms previous methods on both recognition and localization tasks.
A Dataset Generation Framework for Evaluating Megapixel Image Classifiers & Their Explanations
PDF ↗Deep learning-based megapixel image classifiers have exceptional prediction performance in a number of domains, including clinical pathology. However, extracting reliable, human-interpretable model explanations has remained challenging. Because real-world megapixel images often contain latent image features highly correlated with image labels, it is difficult to distinguish correct explanations from incorrect ones. Furthering this issue are the flawed assumptions and designs of today’s classifiers. To investigate classification and explanation performance, we introduce a framework to (a) generate synthetic control images that reflect common properties of megapixel images and (b) evaluate average test-set correctness. By benchmarking two commonplace Convolutional Neural Networks (CNNs), we demonstrate how this interpretability evaluation framework can inform architecture selection beyond classification performance -- in particular, we show that a simple Attention-based architecture identifies salient objects in all seven scenarios, while a standard CNN fails to do so in six scenarios. This work carries widespread applicability to any megapixel imaging domain.
Existing prototypical-based models address the black-box nature of deep learning. However, they are sub-optimal as they often assume separate prototypes for each class, require multi-step optimization, make decisions based on prototype absence (so-called negative reasoning process), and derive vague prototypes. To address those shortcomings, we introduce ProtoPool, an interpretable prototype-based model with positive reasoning and three main novelties. Firstly, we reuse prototypes in classes, which significantly decreases their number. Secondly, we allow automatic, fully differentiable assignment of prototypes to classes, which substantially simplifies the training process. Finally, we propose a new focal similarity function that contrasts the prototype from the background and consequently concentrates on more salient visual features. We show that ProtoPool obtains state-of-the-art accuracy on the CUB-200-2011 and the Stanford Cars datasets, substantially reducing the number of prototypes. We provide a theoretical analysis of the method and a user study to show that our prototypes capture more salient features than those obtained with competitive methods. We made the code available at https://github.com/gmum/ProtoPool.
As AI technology is increasingly applied to high-impact, high-risk domains, there have been a number of new methods aimed at making AI models more human interpretable. Despite the recent growth of interpretability work, there is a lack of systematic evaluation of proposed techniques. In this work, we introduce HIVE (Human Interpretability of Visual Explanations), a novel human evaluation framework that assesses the utility of explanations to human users in AI-assisted decision making scenarios, and enables falsifiable hypothesis testing, cross-method comparison, and human-centered evaluation of visual interpretability methods. To the best of our knowledge, this is the first work of its kind. Using HIVE, we conduct IRB-approved human studies with nearly 1000 participants and evaluate four methods that represent the diversity of computer vision interpretability works: GradCAM, BagNet, ProtoPNet, and ProtoTree. Our results suggest that explanations engender human trust, even for incorrect predictions, yet are not distinct enough for users to distinguish between correct and incorrect predictions. We open-source HIVE to enable future studies and encourage more human-centered approaches to interpretability research.
Rolling shutter (RS) distortion can be interpreted as the result of picking a row of pixels from instant global shutter (GS) frames over time during the exposure of the RS camera. This means that the information of each instant GS frame is partially, yet sequentially, embedded into the row-dependent distortion. Inspired by this fact, we address the challenging task of reversing this process, i.e., extracting undistorted GS frames from images suffering from RS distortion. However, since RS distortion is coupled with other factors such as readout settings and the relative velocity of scene elements to the camera, models that only exploit the geometric correlation between temporally adjacent images suffer from poor generality in processing data with different readout settings and dynamic scenes with both camera motion and object motion. In this paper, instead of two consecutive frames, we propose to exploit a pair of images captured by dual RS cameras with reversed RS directions for this highly challenging task. Grounded on the symmetric and complementary nature of dual reversed distortion, we develop a novel end-to-end model, IFED, to generate dual optical flow sequence through iterative learning of the velocity field during the RS time. Extensive experimental results demonstrate that IFED is superior to naive cascade schemes, as well as the state-of-the-art which utilizes adjacent RS images. Most importantly, although it is trained on a synthetic dataset, IFED is shown to be effective at retrieving GS frame sequences from real-world RS distorted images of dynamic scenes. Code is available at https://github.com/zzh-tech/Dual-Reversed-RS.
We present a method for estimating lighting from a single perspective image of an indoor scene. Previous methods for predicting indoor illumination usually focus on either simple, parametric lighting that lack realism, or on richer representations that are difficult or even impossible to understand or modify after prediction. We propose a pipeline that estimates a parametric light that is easy to edit and allows renderings with strong shadows, alongside with a non-parametric texture with high-frequency information necessary for realistic rendering of specular objects. Once estimated, the predictions obtained with our model are interpretable and can easily be modified by an artist/user with a few mouse clicks. Quantitative and qualitative results show that our approach makes indoor lighting estimation easier to handle by a casual user, while still producing competitive results.
Connectionist Temporal Classification (CTC) is a popular objective function in sequence recognition, which provides supervision for unsegmented sequence data through aligning sequence and its corresponding labeling iteratively. The blank class of CTC plays a crucial role in the alignment process and is often considered responsible for the peaky behavior of CTC. In this study, we propose an objective function named RadialCTC that constrains sequence features on a hypersphere while retaining the iterative alignment mechanism of CTC. The learned features of each non-blank class are distributed on a radial arc from the center of the blank class, which provides a clear geometric interpretation and makes the alignment process more efficient. Besides, RadialCTC can control the peaky behavior by simply modifying the logit of the blank class. Experimental results of recognition and localization demonstrate the effectiveness of RadialCTC on two sequence recognition applications.
Deep implicit functions have shown remarkable shape modeling ability in various 3D computer vision tasks. One drawback is that it is hard for them to represent a 3D shape as multiple parts. Current solutions learn various primitives and blend the primitives directly in the spatial space, which still struggle to approximate the 3D shape accurately. To resolve this problem, we introduce a novel implicit representation to represent a single 3D shape as a set of parts in the latent space, towards both highly accurate and plausibly interpretable shape modeling. Our insight here is that both the part learning and the part blending can be conducted much easier in the latent space than in the spatial space. We name our method Latent Partition Implicit (LPI), because of its ability of casting the global shape modeling into multiple local part modeling, which partitions the global shape unity. LPI represents a shape as Signed Distance Functions (SDFs) using surface codes. Each surface code is a latent code representing a part whose center is on the surface, which enables us to flexibly employ intrinsic attributes of shapes or additional surface properties. Eventually, LPI can reconstruct both the shape and the parts on the shape, both of which are plausible meshes. LPI is a multi-level representation, which can partition a shape into different numbers of parts after training. LPI can be learned without ground truth signed distances, point normals or any supervision for part partition. LPI outperforms the state-of-the-art methods under the widely used benchmarks in terms of reconstruction accuracy and modeling interpretability.
Sketch-and-extrude is a common and intuitive modeling process in computer aided design. This paper studies the problem of learning the shape given in the form of point clouds by “inverse” sketch-and-extrude. We present ExtrudeNet, an unsupervised end-to-end network for discovering sketch and extrude from point clouds. Behind ExtrudeNet are two new technical components: 1) the use of a specially-designed rational Bézier representation for sketch and extrude, which can model extrusion with freeform sketches and conventional cylinder and box primitives as well; and 2) a numerical method for computing the signed distance field which is used in the network learning. This is the first attempt that uses machine learning to reverse engineer the sketch-and-extrude modeling process of a shape in an unsupervised fashion. ExtrudeNet not only outputs a compact, editable and interpretable representation of the shape that can be seamlessly integrated into modern CAD software, but also aligns with the standard CAD modeling process facilitating various editing applications, which distinguishes our work from existing shape parsing research. Code will be open-sourced upon acceptance.
Visual re-localization aims to recover camera poses in a known environment, which is vital for applications like robotics or augmented reality. Feed-forward absolute camera pose regression methods directly output poses by a network, but suffer from low accuracy. Meanwhile, scene coordinate based methods are accurate, but need iterative RANSAC post-processing, which brings challenges to efficient end-to-end training and inference. In order to have the best of both worlds, we propose a feed-forward method termed SC-wLS that exploits all scene coordinate estimates for weighted least squares pose regression. This differentiable formulation exploits a weight network imposed on 2D-3D correspondences, and requires pose supervision only. Qualitative results demonstrate the interpretability of learned weights. Evaluations on 7Scenes and Cambridge datasets show significantly promoted performance when compared with former feed-forward counterparts. Moreover, our SC-wLS method enables a new capability: self-supervised test-time adaptation on the weight network. Codes and models are publicly available.
Self-supervised visual representation learning has achieved remarkable success in recent years. By subverting the need for supervised labels, such approaches are able to utilize the numerous unlabeled images that exist on the Internet and in photographic datasets. Yet to build truly intelligent agents, we must construct representation learning algorithms that can learn not only from datasets but also learn in environments. An agent in a natural environment will not typically be fed curated data. Instead, it must explore its environment to acquire the data it will learn from. We propose a framework, curious representation learning (CRL), which jointly learns a reinforcement learning policy and a visual representation model. The policy is trained to maximize the error of the representation learner, and in doing so is incentivized to explore its environment. At the same time, the learned representation becomes stronger and stronger as the policy feeds it ever harder data to learn from. Our learned embodied representations enable promising transfer to downstream embodied semantic and language-guided navigation, performing better or comparable to ImageNet pretraining without using any supervision at all. In addition, despite being trained in simulation, our learned representations can obtain interpretable results on real images.
Monocular 3D object detection is of great significance for autonomous driving but remains challenging. The core challenge is to predict the distance of objects in the absence of explicit depth information. Unlike regressing the distance as a single variable in most existing methods, we propose a novel geometry-based distance decomposition to recover the distance by its factors. The decomposition factors the distance of objects into the most representative and stable variables, i.e. the physical height and the projected visual height in the image plane. Moreover, the decomposition maintains the self-consistency between the two heights, leading to robust distance prediction when both predicted heights are inaccurate. The decomposition also enables us to trace the causes of the distance uncertainty for different scenarios. Such decomposition makes the distance prediction interpretable, accurate, and robust. Our method directly predicts 3D bounding boxes from RGB images with a compact architecture, making the training and inference simple and efficient. The experimental results show that our method achieves the state-of-the-art performance on the monocular 3D Object Detection and Bird's Eye View tasks of the KITTI dataset, and can generalize to images with different camera intrinsics.
Knowledge Distillation has shown very promising ability in transferring learned representation from the larger model (teacher) to the smaller one (student). Despite many efforts, prior methods ignore the important role of retaining inter-channel correlation of features, leading to the lack of capturing intrinsic distribution of the feature space and sufficient diversity properties of features in the teacher network. To solve the issue, we propose the novel Inter-Channel Correlation for Knowledge Distillation (ICKD), with which the diversity and homology of the feature space of the student network can align with that of the teacher network. The correlation between these two channels is interpreted as diversity if they are irrelevant to each other, otherwise homology. Then the student is required to mimic the correlation within its own embedding space. In addition, we introduce the grid-level inter-channel correlation, making it capable of dense prediction tasks. Extensive experiments on two vision tasks, including ImageNet classification and Pascal VOC segmentation, demonstrate the superiority of our ICKD, which consistently outperforms many existing methods, advancing the state-of-the-art in the fields of Knowledge Distillation. To our knowledge, we are the first method based on knowledge distillation boosts ResNet18 beyond 72% Top-1 accuracy on ImageNet classification. Code is available at: https://github.com/ADLab-AutoDrive/ICKD.
Deformable templates are essential to large-scale medical image registration, segmentation, and population analysis. Current conventional and deep network-based methods for template construction use only regularized registration objectives and often yield templates with blurry and/or anatomically implausible appearance, confounding downstream biomedical interpretation. We reformulate deformable registration and conditional template estimation as an adversarial game wherein we encourage realism in the moved templates with a generative adversarial registration framework conditioned on flexible image covariates. The resulting templates exhibit significant gain in specificity to attributes such as age and disease, better fit underlying group-wise spatiotemporal trends, and achieve improved sharpness and centrality. These improvements enable more accurate population modeling with diverse covariates for standardized downstream analyses and easier anatomical delineation for structures of interest.
We propose three novel solvers for estimating the relative pose of a multi-camera system from affine correspondences (ACs). A new constraint is derived interpreting the relationship of ACs and the generalized camera model. Using the constraint, we demonstrate efficient solvers for two types of motions assumed. Considering that the cameras undergo planar motion, we propose a minimal solution using a single AC and a solver with two ACs to overcome the degenerate case. Also, we propose a minimal solution using two ACs with known vertical direction, e.g., from an IMU. Since the proposed methods require significantly fewer correspondences than state-of-the-art algorithms, they can be efficiently used within RANSAC for outlier removal and initial motion estimation. The solvers are tested both on synthetic data and on real-world scenes from the KITTI odometry benchmark. It is shown that the accuracy of the estimated poses is superior to the state-of-the-art techniques.
Colorization has attracted increasing interest in recent years. Classic reference-based methods usually rely on external color images for plausible results. A large image database or online search engine is inevitably required for retrieving such exemplars. Recent deep-learning-based methods could automatically colorize images at a low cost. However, unsatisfactory artifacts and incoherent colors are always accompanied. In this work, we aim at recovering vivid colors by leveraging the rich and diverse color priors encapsulated in a pretrained Generative Adversarial Networks (GAN). Specifically, we first "retrieve" matched features (similar to exemplars) via a GAN encoder and then incorporate these features into the colorization process with feature modulations. Thanks to the powerful generative color prior and delicate designs, our method could produce vivid colors with a single forward pass. Moreover, it is highly convenient to obtain diverse results by modifying GAN latent codes. Our method also inherits the merit of interpretable controls of GANs and could attain controllable and smooth transitions by walking through GAN latent space. Extensive experiments and user studies demonstrate that our method achieves superior performance than previous works.
Deep neural networks achieve great success in many visual recognition tasks. However, the model deployment is usually subject to some computational resources. Model pruning under computational budget has attracted growing attention. In this paper, we focus on the discrimination-aware compression of Convolutional Neural Networks (CNNs). In prior arts, directly searching the optimal sub-network is an integer programming problem, which is non-smooth, non-convex, and NP-hard. Meanwhile, the heuristic pruning criterion lacks clear interpretability and doesn't generalize well in applications. To address this problem, we formulate sub-networks as samples from a multivariate Bernoulli distribution and resort to the approximation of continuous problem. We propose a new flexible search scheme via alternating exploration and estimation. In the exploration step, we employ stochastic gradient Hamiltonian Monte Carlo with budget-awareness to generate sub-networks, which allows large search space with efficient computation. In the estimation step, we deduce the sub-network sampler to a near-optimal point, to promote the generation of high-quality sub-networks. Unifying the exploration and estimation, our approach avoids early falling into local minimum via a fast gradient-based search in a larger space. Extensive experiments on CIFAR-10 and ImageNet show that our method achieves state-of-the-art performances on pruning several popular CNNs.
This paper aims to explain adversarial attacks in terms of how adversarial perturbations contribute to the attacking task. We estimate attributions of different image regions to the decrease of the attacking cost based on the Shapley value. We define and quantify interactions among adversarial perturbation pixels, and decompose the entire perturbation map into relatively independent perturbation components. The decomposition of the perturbation map shows that adversarially-trained DNNs have more perturbation components in the foreground than normally-trained DNNs. Moreover, compared to the normally-trained DNN, the adversarially-trained DNN have more components which mainly decrease the score of the true category. Above analyses provide new insights into the understanding of adversarial attacks.