ECCV 2020 Papers

We formulate counting as a sequential decision problem and present a novel crowd counting model solvable by deep reinforcement learning. In contrast to existing counting models that directly output count values, we divide one-step estimation into a sequence of much easier and more tractable sub-decision problems. Such sequential decision nature corresponds exactly to a physical process in reality|scale weighing. Inspired by scale weighing, we propose a novel ‘counting scale’ termed LibraNet where the count value is analogized by weight. By virtually placing a crowd image on one side of a scale, LibraNet (agent) sequentially learns to place appropriate weights on the other side to match the crowd count. At each step, LibraNet chooses one weight (action) from the weight box (the pre-defined action pool) according to the current crowd image features and weights placed on the scale pan (state). LibraNet is required to learn to balance the scale according to the feedback of the needle (Q values). We show that LibraNet exactly implements scale weighing by visualizing the decision process how LibraNet chooses actions. Extensive experiments demonstrate the effectiveness of our design choices and report state-of-the-art results on a few crowd counting benchmarks, including ShanghaiTech, UCF_CC_50 and UCF-QNRF. We also demonstrate good cross-dataset generalization of LibraNet. Code and models are made available at https://git.io/libranet"

This paper focuses on webly supervised learning (WSL), where datasets are built by crawling samples from the Internet and adopting search queries directly as their web labels. Although WSL benefits from fast and low-cost data expansion, noisy web labels prevent models from reliable predictions. To mitigate this problem, self-labeled supervised loss $\mathcal{L}_s$ is utilized together with webly supervised loss $\mathcal{L}_w$ in recent works. $\mathcal{L}_s$ relies on machine labels predicted by the model itself. Since whether web labels or machine labels being correct is usually on a case-by-case basis for web samples, it is desirable to adjust balance between $\mathcal{L}_s$ and $\mathcal{L}_w$ at a sample-level. Inspired by DNNs' ability on confidence prediction, we introduce self-contained confidence (SCC) by adapting model uncertainty for WSL setting and use it to sample-wisely balance $\mathcal{L}_s$ and $\mathcal{L}_w$. Thus, a simple yet effective WSL framework is proposed. A series of SCC-friendly approaches are investigated, among which our proposed graph-enhanced mixup stands out as the most effective approach to provide high-quality confidence to boost our framework. The proposed WSL framework achieved state-of-the-art results on two large-scale WSL datasets of WebVision-1000 and Food101-N."

This paper proposes a weakly-supervised learning framework for dynamics estimation from human motion. Although there are many solutions to capture pure human motion readily available, their data is not sufficient to analyze quality and efficiency of movements. Instead, the forces and moments driving human motion (the dynamics) need to be considered. Since recording dynamics is a laborious task that requires expensive sensors and complex, time-consuming optimization, dynamics data sets are small compared to human motion data sets and are rarely made public. The proposed approach takes advantage of easily obtainable motion data which enables weakly-supervised learning on small dynamics sets and weakly-supervised domain transfer. Our method includes novel neural network (NN) layers for forward and inverse dynamics during end-to-end training. On this basis, a cyclic loss between pure motion data can be minimized, i.e. no ground truth forces and moments are required during training. The proposed method achieves state-of-the-art results in terms of ground reaction force, ground reaction moment and joint torque regression and is able to maintain good performance on substantially reduced sets."

3D shape completion for real data is important but challenging, since partial point clouds acquired by real-world sensors are usually sparse, noisy and unaligned. Different from previous methods, we address the problem of learning 3D complete shape from unaligned and real-world partial point clouds. To this end, we propose an unsupervised method to estimate both 3D canonical shape and 6-DoF pose for alignment, given multiple partial observations associated with the same instance. The network jointly optimizes canonical shapes and poses with multi-view geometry constraints during training, and can infer the complete shape given a single partial point cloud. Moreover, learned pose estimation can facilitate partial point cloud registration. Experiments on both synthetic and real data show that it is feasible and promising to learn 3D shape completion through large-scale data without shape and pose supervision."

In crowd counting datasets, the location labels are costly, yet, they are not taken into the evaluation metrics. Besides, existing multi-task approaches employ high-level tasks to improve counting accuracy. This research tendency increases the demand for more annotations. In this paper, we propose a weakly-supervised counting network, which directly regresses the crowd numbers without the location supervision. Moreover, we train the network to count by exploiting the relationship among the images. We propose a soft-label sorting network along with the counting network, which sorts the given images by their crowd numbers. The sorting network drives the shared backbone CNN model to obtain density-sensitive ability explicitly. Therefore, the proposed method improves the counting accuracy by utilizing the information hidden in crowd numbers, rather than learning from extra labels, such as locations and perspectives.We evaluate our proposed method on three crowd counting datasets, and the performance of our method plays favorably against the fully supervised state-of-the-art approaches."

We propose a weakly-supervised cell tracking method that can train a convolutional neural network (CNN) by using only the annotation of ""cell detection"" (i.e., the coordinates of cell positions) without association information, in which cell positions can be easily obtained by nuclear staining. First, we train a co-detection CNN that detects cells in successive frames by using weak-labels. Our key assumption is that the co-detection CNN implicitly learns association in addition to detection. To obtain the association information, we propose a backward-and-forward propagation method that analyzes the correspondence of cell positions in the detection maps output of the co-detection CNN. Experiments demonstrated that the proposed method can match positions by analyzing the co-detection CNN. Even though the method uses only weak supervision, the performance of our method was almost the same as the state-of-the-art supervised method."

Weakly-supervised action localization requires training a model to localize the action segments in the video given only video level action label. It can be solved under the Multiple Instance Learning (MIL) framework, where a bag (video) contains multiple instances (action segments). Since only the bag's label is known, the main challenge is assigning which key instances within the bag to trigger the bag's label. Most previous models use attention-based approaches applying attentions to generate the bag's representation from instances, and then train it via the bag's classification. These models, however, implicitly violate the MIL assumption that instances in negative bags should be uniformly negative. In this work, we explicitly model the key instances assignment as a hidden variable and adopt an Expectation-Maximization (EM) framework. We derive two pseudo-label generation schemes to model the E and M process and iteratively optimize the likelihood lower bound. We show that our EM-MIL approach more accurately models both the learning objective and the MIL assumptions. It achieves state-of-the-art performance on two standard benchmarks, THUMOS14 and ActivityNet1.2."

Weakly supervised semantic segmentation with image-level labels has attracted a lot of attention recently because these labels are already available in most datasets. To obtain semantic segmentation under weak supervision, this paper presents a simple yet effective approach based on the idea of explicitly exploring object boundaries from training images to keep coincidence of segmentation and boundaries. Specifically, we synthesize boundary annotations by exploiting coarse localization maps obtained from CNN classifier, and use annotations to train the proposed network called BENet which further excavates more object boundaries to provide constraints for segmentation. Finally generated pseudo annotations of training images are used to supervise an off-the-shelf segmentation network. We evaluate the proposed method on PASCAL VOC 2012 benchmark and the final results achieve 65.7% and 66.6% mIoU scores on val and test sets respectively, which outperforms previous methods trained under image-level supervision."

In many real-world datasets, like WebVision, the performance of DNN based classier is often limited by the noisy labeled data. To tackle this problem, some image related side information, such as captions and tags, often reveal underlying relationships across images. In this paper, we present an efficient weakly-supervised learning by using a Side Information Network (SINet), which aims to effectively carry out a large scale classication with severely noisy labels. The proposed SINet consists of a visual prototype module and a noise weighting module. The visual prototype module is designed to generate a compact representation for each category by introducing the side information. The noise weighting module aims to estimate the correctness of each noisy image and produce a condence score for image ranking during the training procedure. The propsed SINet can largely alleviate the negative impact of noisy image labels, and is benecial to train a high performance CNN based classier. Besides, we release a ne-grained product dataset called AliProducts, which contains more than 2.5 million noisy web images crawled from the internet by using queries generated from 50,000 fine-grained semantic classes. Extensive experiments on several popular benchmarks (i.e. Webvision, ImageNet and Clothing-1M) and our proposed AliProducts achieve state-of-the-art performance. The SINet has won the rst place in the 5000 category classication task on WebVision Challenge 2019, and outperforms other competitors by a large margin."

Recent approaches for weakly supervised instance segmentations depend on two components: (i) a pseudo label generation model that provides instances which are consistent with a given annotation; and (ii) an instance segmentation model, which is trained in a supervised manner using the pseudo labels as ground-truth. Unlike previous approaches, we explicitly model the uncertainty in the pseudo label generation process using a conditional distribution. The samples drawn from our conditional distribution provide accurate pseudo labels due to the use of semantic class aware unary terms, boundary aware pairwise smoothness terms, and annotation aware higher order terms. Furthermore, we represent the instance segmentation model as an annotation agnostic prediction distribution. In contrast to previous methods, our representation allows us to define a joint probabilistic learning objective that minimizes the dissimilarity between the two distributions. Our approach achieves state of the art results on the PASCAL VOC 2012 data set, outperforming the best baseline by 4.2% mAP@0.5 and 4.8% mAP@0.75."

It is laborious to manually label point cloud data for training high-quality 3D object detectors. This work proposes a weakly supervised approach for 3D object detection, only requiring a small set of weakly annotated scenes, associated with a few precisely labeled object instances. This is achieved by a two-stage architecture design. Stage-1 learns to generate cylindrical object proposals under weak supervision, i.e., only the horizontal centers of objects are click-annotated on bird’s view scenes. Stage-2 learns to refine the cylindrical proposals to get cuboids and confidence scores, using a few well-labeled object instances. Using only 500 weakly annotated scenes and 534 precisely labeled vehicle instances, our method achieves 85∼95% of performance of current top-leading, fully supervised detectors (requiring 3,712 exhaustively and precisely annotated scenes with 15,654 instances). Moreover, with our elaborately designed network architecture, our trained model can be applied as a 3D object annotator, supporting both automatic and active (human-in-the-loop) working modes. The annotations generated by our model can be used to train 3D object detectors achieving 94% of their original performance (with manually labeled data). Our experiments also show our model's potential in boosting performance when given more training data. Above designs make our approach highly practical and introduce new opportunities for learning 3D object detection at reduced annotation cost."

Monocular 3D human pose and shape estimation is challenging due to the many degrees of freedom of the human body and the difficulty to acquire training data for large-scale supervised learning in complex visual scenes where humans with diverse shape and appearance, appear against complex backgrounds,in a variety of poses, and are partially occluded, or involved in interactions. Essential to learning is leveraging effective 3D human priors, and the ability to work under weak supervision, at scale, by exploiting, to the largest extent, the detailed human body semantics in images. In this paper we present new priors as well as large-scale weakly supervised models for 3D human pose and shape estimation. Key to our formulation are new latent normalizing flow representations, as well as fully differentiable, structurally-sensitive, semantic body part alignment(re-projection) loss functions that ensure consistent estimates and sharp feedback signals for learning. In extensive experiments using both motion capture datasets like CMU, Human3.6M, 3DPW, or AMASS, and repositories like COCO, we show that our proposed methods outperform existing counterparts, supporting the construction of an increasingly more accurate family of models based on large-scale training with unlabeled image data. "

Estimating 3D hand pose from 2D images is a difficult, inverse problem due to the inherent scale and depth ambiguities. Current state-of-the-art methods train fully supervised deep neural networks with 3D ground-truth data. However, acquiring 3D annotations is expensive, typically requiring calibrated multi-view setups or labour intensive manual annotations. While annotations of 2D keypoints are much easier to obtain, how to efficiently leverage such extit{weakly-supervised} data to improve the task of 3D hand pose prediction remains an important open question. The key difficulty stems from the fact that direct application of additional 2D supervision mostly benefits the 2D proxy objective but does little to alleviate the depth and scale ambiguities. Embracing this challenge we propose a set of novel losses that constrain the prediction of a neural network to lie within the range of biomechanically feasible 3D hand configurations. We show by extensive experiments that our proposed constraints significantly reduce the depth ambiguity and allow the network to more effectively leverage additional 2D annotated images. For example, on the challenging freiHAND dataset, using additional 2D annotation without our proposed biomechanical constraints reduces the depth error by only $15\%$, whereas the error is reduced significantly by $50\%$ when the proposed biomechanical constraints are used. "

Identifying common patterns among events is a key capability for human and machine perception, as it underlies intelligent decision making. Here, we propose an approach for learning semantic relational set abstractions on videos, inspired by human learning. Our model combines visual features as input with natural language supervision to generate high-level representations of similarities across a set of videos. This allows our model to perform cognitive tasks such as set abstraction (what is in common among a set of videos?), set completion (which new video goes well with the set?), and odd one out detection (which video does not belong in the set?). Experiments on two video benchmarks, Kinetics and Multi-Moments in Time, show that robust and versatile representations emerge when learning to recognize commonalities among sets. We compare our model to several baseline algorithms and show that significant improvements result from explicitly learning relational abstractions with semantic supervision."

When smartphone cameras are used to take photos of digital screens, usually moire patterns result, severely degrading photo quality. In this paper, we design a wavelet-based dual-branch network (WDNet) with a spatial attention mechanism for image demoireing. Existing image restoration methods working in the RGB domain have difficulty in distinguishing moire patterns from true scene texture. Unlike these methods,our network removes moire patterns in the wavelet domain to separate the frequencies of moire patterns from the image content. The network combines dense convolution modules and dilated convolution modules supporting large receptive fields. Extensive experiments demonstrate the effectiveness of our method, and we further show that WDNet generalizes to removing moire artifacts on non-screen images. Although designed for image demoireing, WDNet has been applied to two other low-level vision tasks, outperforming state-of-the-art image deraining and derain-drop methods on the Rain100h and Raindrop800 data sets, respectively."

We present mph{VoxelPose} to estimate $3$D poses of multiple people from multiple camera views. In contrast to the previous efforts which require to establish cross-view correspondence based on noisy and incomplete $2$D pose estimates, mph{VoxelPose} directly operates in the $3$D space therefore avoids making incorrect decisions in each camera view. To achieve this goal, features in all camera views are aggregated in the $3$D voxel space and fed into mph{Cuboid Proposal Network} (CPN) to localize all people. Then we propose mph{Pose Regression Network} (PRN) to estimate a detailed $3$D pose for each proposal. The approach is robust to occlusion which occurs frequently in practice. Without bells and whistles, it outperforms the previous methods on several public datasets."

Existing techniques to encode spatial invariance within deep convolutional neural networks (CNNs) apply the same warping field to all the feature channels. This does not account for the fact that the individual feature channels can represent different semantic parts, which can undergo different spatial transformations w.r.t. a canonical configuration. To overcome this limitation, we introduce a learnable module, the volumetric transformer network (VTN), that predicts channel-wise warping fields so as to reconfigure intermediate CNN features spatially and channel-wisely. We design our VTN as an encoder-decoder network, with modules dedicated to letting the information flow across the feature channels, to account for the dependencies between the semantic parts. We further propose a loss function defined between the warped features of pairs of instances, which improves the localization ability of VTN. Our experiments show that VTN consistently boosts the features' representation power and consequently the networks' accuracy on fine-grained image recognition and instance-level image retrieval."

Several animal species (e.g., bats, dolphins, and whales) and even visually impaired humans have the remarkable ability to perform echolocation: a biological sonar used to perceive spatial layout and locate objects in the world. We explore the spatial cues contained in echoes and how they can benefit vision tasks that require spatial reasoning. First we capture echo responses in photo-realistic 3D indoor scene environments. Then we propose a novel interaction-based representation learning framework that learns useful visual features via echolocation. We show that the learned image features are useful for multiple downstream vision tasks requiring spatial reasoning---monocular depth estimation, surface normal estimation, and visual navigation---with results comparable or even better than heavily supervised pre-training. Our work opens a new path for representation learning for embodied agents, where supervision comes from interacting with the physical world."

Even from a single frame of a still image, people can reason about the dynamic story of the image before, after, and beyond the frame. For example, given an image of a man struggling to stay afloat in water, we can reason that the man fell into the water sometime in the past, the intent of that man at the moment is to stay alive, and he will need help in the near future or else he will get washed away. We propose Visual COMET, the novel framework of visual common-sense reasoning tasks to predict events that might have happened before, events that might happen next, and the intents of the people at present. To support research toward visual commonsense reasoning, we introduce the first large-scale repository of Visual Commonsense Graphs that consists of over 1.4 million textual descriptions of visual commonsense inferences carefully annotated over a diverse set of 59,000 images, each paired with short video summaries of before and after. In addition, we provide person-grounding (i.e., co-reference links) between people appearing in the image and people mentioned in the textual commonsense descriptions, allowing for tighter integration between images and text. We establish strong baseline performances on this task and demonstrate that integration between visual and textual commonsense reasoning is the key and wins over non-integrative alternatives."