WACV 2021 Papers

We propose an optimal transport (OT) framework for generalized zero-shot learning (GZSL), seeking to distinguish samples for both seen and unseen classes, with the assist of auxiliary attributes. The discrepancy between features and attributes is minimized by solving an optimal transport problem. Specifically, we build a conditional generative model to generate features from seen-class attributes, and establish an optimal transport between the distribution of the generated features and that of the real features. The generative model and the optimal transport are optimized iteratively with an attribute-based regularizer, that further enhances the discriminative power of the generated features. A classifier is learned based on the features generated for both the seen and unseen classes. In addition to generalized zero-shot learning, our framework is also applicable to standard and transductive ZSL problems. Experiments show that our optimal transport-based method outperforms state-of-the-art methods on several benchmark datasets.

In this paper, we propose an approach based on domain conditional normalization (DCN) for zero-pair image-to-image translation, i.e., translating between two domains which have no paired training data available but each have paired training data with a third domain. We employ a single generator which has an encoder-decoder structure and analyze different implementations of domain conditional normalization to obtain the desired target domain output. The validation benchmark uses RGB-depth pairs and RGB-semantic pairs for training and compares performance for the depth-semantic translation task. The proposed approaches improve in qualitative and quantitative terms over the compared methods, while using much fewer parameters.

Recognizing people by faces and other biometrics has been extensively studied in computer vision. But these techniques do not work for identifying the wearer of an egocentric (first-person) camera because that person rarely (if ever) appears in their own first-person view. But while one's own face is not frequently visible, their hands are: in fact, hands are among the most common objects in one's own field of view. It is thus natural to ask whether the appearance and motion patterns of people's hands are distinctive enough to recognize them. In this paper, we systematically study the possibility of Egocentric Hand Identification (EHI) with unconstrained egocentric hand gestures. We explore several different visual cues, including color, shape, skin texture, and depth maps to identify users' hands. Extensive ablation experiments are conducted to analyze the properties of hands that are most distinctive. Finally, we show that EHI can improve generalization of other tasks, such as gesture recognition, by training adversarially to encourage these models to ignore differences between users.

Humans possess an innate capability of recognizing objects and their corresponding parts and confine their attention to that location in a visual scene where the object is spatially present. Recently, efforts to train machines to mimic this ability of humans in the form of weakly supervised object localization, using training labels only at the image-level, have garnered a lot of attention. Nonetheless, one of the well-known problems that most of the existing methods suffer from is localizing only the most discriminative part of an object. Such methods provide very little or no focus on other pertinent parts of the object. In this paper, we propose a novel way of scrupulously localizing objects using training with labels as for the entire image by mining information from complementary regions in an image. Primarily, we adapt to regional dropout at complementary spatial locations to create two intermediate images. With the help of a novel Channel-wise Assisted Attention Module (CAAM) coupled with a Spatial Self-Attention Module (SSAM), we parallely train our model to leverage the information from complementary image regions for excellent localization. Finally, we fuse the attention maps generated by the two classifiers using our Attention-based Fusion Loss. Several experimental studies manifest the superior performance of our proposed approach. Our method demonstrates a significant increase in localization performance over the existing state-of-the-art methods on CUB-200-2011 and ILSVRC 2016 datasets.

Training a semantic segmentation model requires large densely-annotated image datasets that are costly to obtain. Once the training is done, it is also difficult to add new object categories to such segmentation models. In this paper, we tackle the few-shot semantic segmentation problem, which aims to perform image segmentation task on unseen object categories merely based on one or a few support example(s). The key to solving this few-shot segmentation problem lies in effectively utilizing object information from support examples to separate target objects from the background in a query image. While existing methods typically generate object-level representations by averaging local features in support images, we demonstrate that such object representations are typically noisy and less distinguishing. To solve this problem, we design an object representation generator (ORG) module which can effectively aggregate local object features from support image(s) and produce better object-level representation. The ORG module can be embedded into the network and trained end-to-end in a weakly-supervised fashion without extra human annotation. We incorporate this design into a modified encoder-decoder network to present a powerful and efficient framework for few-shot semantic segmentation. Experimental results on the Pascal-VOC and MS-COCO datasets show that our approach achieves better performance compared to existing methods under both one-shot and five-shot settings.

We present a weakly supervised instance segmentation algorithm based on deep community learning with multiple tasks. This task is formulated as a combination of weakly supervised object detection and semantic segmentation, where individual objects of the same class are identified and segmented separately. We address this problem by designing a unified deep neural network architecture, which has a positive feedback loop of object detection with bounding box regression, instance mask generation, instance segmentation, and feature extraction. Each component of the network makes active interactions with others to improve accuracy, and the end-to-end trainability of our model makes our results more robust and reproducible. The proposed algorithm achieves state-of-the-art performance in the weakly supervised setting without any additional training such as Fast R-CNN and Mask R-CNN on the standard benchmark dataset. The implementation of our algorithm is available on the project webpage: https://cv.snu.ac.kr/research/WSIS_CL.

Conventional unsupervised video summarization algorithms are usually developed in a frame level clustering manner. For example, frame level diversity and representativeness are two typical clustering criteria used for unsupervised reinforcement learning-based video summarization. Inspired by recent progress in video representation techniques, we further introduce the similarity of video representations to construct a semantically meaningful reward for this task. We consider that a good summarization should also be semantically identical to its original source, which means that the semantic similarity can be regarded as an additional criterion for summarization. Through combining a novel video semantic reward with other unsupervised rewards for training, we can easily upgrade an unsupervised reinforcement learning-based video summarization method to its weakly supervised version. In practice, we first train a video classification sub-network (VCSN) to extract video semantic representations based on a category-labeled video dataset. Then we fix this VCSN and train a summary generation sub-network (SGSN) using unlabeled video data in a reinforcement learning way. Experimental results demonstrate that our work significantly surpasses other unsupervised and even supervised methods. To the best of our knowledge, our method achieves state-of-the-art performance in terms of the correlation coefficients, Kendall's \tau and Spearman's \rho.

We propose SSA2D, a simple yet effective end-to-end deep network for actor-action detection in videos. The existing methods take a top-down approach based on region-proposals (RPN), where the action is estimated based on the detected proposals followed by post-processing such as non-maximal suppression. While effective in terms of performance, these methods pose limitations in scalability for dense video scenes with a high memory requirement for thousands of proposals. We propose to solve this problem from a different perspective where we don't need any proposals. SSA2D is a unified network, which performs pixel level joint actor-action detection in a single-shot, where every pixel of the detected actor is assigned an action label. SSA2D has two main advantages: 1) It is a fully convolutional network which does not require any proposals and post-processing making it memory as well as time efficient, 2) It is easily scalable to dense video scenes as its memory requirement is independent of the number of actors present in the scene. We evaluate the proposed method on the Actor-Action dataset (A2D) and Video Object Relation (VidOR) dataset, demonstrating its effectiveness in multiple actors and action detection in a video. SSA2D is 11x faster during inference with comparable (sometimes better) performance and fewer network parameters when compared with the prior works. Code available at https://github.com/aayushjr/ssa2d

Visible watermarks are widely-used in images to protect copyright ownership. Analyzing watermark removal helps to reinforce the anti-attack techniques in an adversarial way. Current removal methods normally leverage image-to-image translation techniques. Nevertheless, the uncertainty of the size, shape, color and transparency of the watermarks set a huge barrier for these methods. To combat this, we combine traditional watermarked image decomposition into a two-stage generator, called Watermark-Decomposition Network (WDNet), where the first stage predicts a rough decomposition from the whole watermarked image and the second stage specifically centers on the watermarked area to refine the removal results. The decomposition formulation enables WDNet to separate watermarks from the images rather than simply removing them. We further show that these separated watermarks can serve as extra nutrients for building a larger training dataset and further improving removal performance. Besides, we construct a large-scale dataset named CLWD, which mainly contains colored watermarks, to fill the vacuum of colored watermark removal dataset. Extensive experiments on the public gray-scale dataset LVW and CLWD consistently show that the proposed WDNet outperforms the state-of-the-art approaches both in accuracy and efficiency.

The ocean is a vast three-dimensional space that is poorly explored and understood, and harbors unobserved life and processes that are vital to ecosystem function. To fully interrogate the space, novel algorithms and robotic platforms are required to scale up observations. Locating animals of interest and extended visual observations in the water column are particularly challenging objectives. Towards that end, we present a novel Machine Learning-integrated Tracking (or ML-Tracking) algorithm for underwater vehicle control that builds on the class of algorithms known as tracking-by-detection. By coupling a multi-object detector (trained on in situ underwater image data), a 3D stereo tracker, and a supervisor module to oversee the mission, we show how ML-Tracking can create robust tracks needed for long duration observations, as well as enable fully automated acquisition of objects for targeted sampling. Using a remotely operated vehicle as a proxy for an autonomous underwater vehicle, we demonstrate continuous input from the ML-Tracking algorithm to the vehicle controller during a record, 5+ hr continuous observation of a midwater gelatinous animal known as a siphonophore. These efforts clearly demonstrate the potential that tracking-by-detection algorithms can have on exploration in unexplored environments and discovery of undiscovered life in our ocean.

In this work, we re-think the task of speech enhancement in unconstrained real-world environments. Current state-of-the-art methods use only the audio stream and are limited in their performance in a wide range of real-world noises. Recent works using lip movements as additional cues improve the quality of generated speech over "audio-only" methods. But, these methods cannot be used for several applications where the visual stream is unreliable or completely absent. We propose a new paradigm for speech enhancement by exploiting recent breakthroughs in speech-driven lip synthesis. Using one such model as a teacher network, we train a robust student network to produce accurate lip movements that mask away the noise, thus acting as a "visual noise filter". The intelligibility of the speech enhanced by our pseudo-lip approach is comparable (< 3% difference) to the case of using real lips. This implies that we can exploit the advantages of using lip movements even in the absence of a real video stream. We rigorously evaluate our model using quantitative metrics as well as human evaluations. Additional ablation studies and a demo video on our website containing qualitative comparisons and results clearly illustrate the effectiveness of our approach.

Rendering an any-viewpoint image is extremely difficult for Generative Adversarial Networks. This is because conventional GANs do not understand 3D information underlying a given viewpoint image such as an object shape and relationship between viewpoint and objects in 3D space. In this paper, we present how to perform a Viewpoint-Agnostic Image Rendering (VAIR), equipping a conditional GAN with a mechanism to reconstruct 3D information of the input view. VAIR realizes any-viewpoint image generation by manipulating a viewpoint in 3D space where the reconstructed instance shape is arranged. In addition, we convert the reconstructed 3D shape into a 2D representation for image-based conditional GAN, while preserving detail 3D information. The representation consists of a depth image and 2D semantic keypoint images, which are obtained by rendering the shape from a viewpoint. In the experiment, we evaluate using a CUB-200-2011 dataset, which contains few-samples biased a viewpoint such that covers only part of the target appearance. As a result, our VAIR clearly renders an any-viewpoint image.

We propose a semi-supervised learning approach for video classification, VideoSSL, using convolutional neural networks (CNN). Like other computer vision tasks, existing supervised video classification methods demand a large amount of labeled data to attain good performance. However, annotation of a large dataset is expensive and time consuming. To minimize the dependence on a large annotated dataset, our proposed semi-supervised method trains from a small number of labeled examples and exploits two regulatory signals from unlabeled data. The first signal is the pseudo-labels of unlabeled examples computed from the confidences of the CNN being trained. The other is the normalized probabilities, as predicted by an image classifier CNN, that captures the information about appearances of the interesting objects in the video. We show that, under the supervision of these guiding signals from unlabeled examples, a video classification CNN can achieve impressive performances utilizing a small fraction of annotated examples on three publicly available datasets: UCF101, HMDB51 and Kinetics.

Being able to anticipate and describe what may happen in the future is a fundamental ability for humans. Given a short clip of a scene about "a person is sitting behind a piano", humans can describe what will happen afterward, i.e. "the person is playing the piano". In this paper, we consider the task of captioning future events to assess the performance of intelligent models on anticipation and video description generation tasks simultaneously. More specifically, given only the frames relating to an occurring event (activity), the goal is to generate a sentence describing the most likely next event in the video. We tackle the problem by first predicting the next event in the semantic space of convolutional features, then fusing contextual information into those features, and feeding them to a captioning module. Departing from using recurrent units allows us to train the network in parallel. We compare the proposed method with a baseline and an oracle method on the ActivityNetCaptions dataset. Experimental results demonstrate that the proposed method outperforms the baseline and is comparable to the oracle method. We perform additional ablation study to further analyze our approach.

Detecting subtle intention such as deception and subtext of a person in a long dialog video, or implicit intention detection (IID), is a challenging problem. The transcript (textual cues) often reveals little, so audio-visual cues including voice tone as well as facial and body behaviour are the main focuses for automated IID. Contextual cues are also crucial, since a person's implicit intentions are often correlated and context-dependent when the person moves from one question-answer pair to the next. However, no such dataset exists which contains fine-grained question-answer pair (video segment) level annotation. The first contribution of this work is thus a new benchmark dataset, called Vid2Int-Deception to fill this gap. A novel multi-grain representation model is also proposed to capture the subtle movement changes of eyes, face, and body (relevant for inferring intention) from a long dialog video. Moreover, to model the temporal correlation between the implicit intentions across video segments, we propose a Video-to-Intention network (Vid2Int) based on attentive recurrent neural network (RNN). Extensive experiments show that our model achieves state-of-the-art results.

In this paper, we propose variational prototype inference to address few-shot semantic segmentation in a probabilistic framework. A probabilistic latent variable model infers the distribution of the prototype that is treated as the latent variable. We formulate the optimization as a variational inference problem, which is established with an amortized inference network based on an auto-encoder architecture. The probabilistic modeling of the prototype enhances its generalization ability to handle the inherent uncertainty caused by limited data and the huge intra-class variations of objects. Moreover, it offers a principled way to incorporate the prototype extracted from support images into the prediction of the segmentation maps for query images. We conduct extensive experimental evaluation on three benchmark datasets. Ablation studies show the effectiveness of variational prototype inference for few-shot semantic segmentation by probabilistic modeling. On all three benchmarks, our proposal achieves high segmentation accuracy and surpasses previous methods by considerable margins.

Phrase grounding models localize an object in the image given a referring expression. The annotated language queries available during training are limited, which also limits the variations of language combinations that a model can see during training. In this paper, we study the case applying objects without labeled queries for training the semi-supervised phrase grounding. We propose to use learned location and subject embedding predictors (LSEP) to generate the corresponding language embeddings for objects lacking annotated queries in the training set. With the assistance of the detector, we also apply LSEP to train a grounding model on images without any annotation. We evaluate our method based on MAttNet on three public datasets: RefCOCO, RefCOCO+, and RefCOCOg. We show that our predictors allow the grounding system to learn from the objects without labeled queries and improve accuracy by 34.9% relatively with the detection results.

This paper introduces a novel method for self-supervised video representation learning via feature prediction. In contrast to the previous methods that focus on future feature prediction, we argue that a supervisory signal arising from unobserved past frames is complementary to one that originates from the future frames. The rationale behind our method is to encourage the network to explore the temporal structure of videos by distinguishing between future and past given present observations. We train our model in a contrastive learning framework, where joint encoding of future and past provides us with a comprehensive set of temporal hard negatives via swapping. We empirically show that utilizing both signals enriches the learned representations for the downstream task of action recognition. It outperforms independent prediction of future and past.

Existing unsupervised video-to-video translation methods fail to produce translated videos which are frame-wise realistic, semantic information preserving and video-level consistent. In this work, we propose a novel unsupervised video-to-video translation model. Our model decomposes the style and the content uses the specialized encoder-decoder structure and propagates the inter-frame information through bidirectional recurrent neural network (RNN) units. The style-content decomposition mechanism enables us to achieve style-consistent video translation results as well as provides us with a good interface for modality flexible translation. In addition, by changing the input frames and style codes incorporated in our translation, we propose a video interpolation loss, which captures temporal information within the sequence to train our building blocks in a self-supervised manner. Our model can produce photo-realistic, spatio-temporal consistent translated videos in a multimodal way. Subjective and objective experimental results validate the superiority of our model over existing methods.