Visual Question Answering (VQA) is a challenging task for evaluating the ability of comprehensive understanding of the world. Existing benchmarks usually focus on the reasoning abilities either only on the vision or mainly on the knowledge with relatively simple abilities on vision. However, the ability of answering a question that requires alternatively inferring on the image content and the commonsense knowledge is crucial for an advanced VQA system. In this paper, we introduce a VQA dataset that provides more challenging and general questions about Compositional Reasoning on vIsion and Commonsense, which is named as CRIC. To create this dataset, we develop a powerful method to automatically generate compositional questions and rich annotations from both the scene graph of a given image and some external knowledge graph. Moreover, this paper presents a new compositional model that is capable of implementing various types of reasoning functions on the image content and the knowledge graph. Further, we analyze several baselines, state-of-the-art and our model on CRIC dataset. The experimental results show that the proposed task is challenging, where state-of-the-art obtains 52.26% accuracy and our model obtains 58.38%.
Zero-shot learning (ZSL) is a challenging problem that aims to recognize the target categories without seen data, where semantic information is leveraged to transfer knowledge from some source classes. Although ZSL has made great progress in recent years, most existing approaches are easy to overfit the sources classes in generalized zero-shot learning (GZSL) task, which indicates that they learn little knowledge about target classes. To tackle such problem, we propose a novel Transferable Contrastive Network (TCN) that explicitly transfers knowledge from the source classes to the target classes. It automatically contrasts one image with different classes to judge whether they are consistent or not. By exploiting the class similarities to make knowledge transfer from source images to similar target classes, our approach is more robust to recognize the target images. Experiments on five benchmark datasets show the superiority of our approach for GZSL.
In many scenarios of Person Re-identification (Re-ID), the gallery set consists of lots of surveillance videos and the query is just an image, thus Re-ID has to be conducted between image and videos. Compared with videos, still person images lack temporal information. Besides, the information asymmetry between image and video features increases the difficulty in matching images and videos. To solve this problem, we propose a novel Temporal Knowledge Propagation (TKP) method which propagates the temporal knowledge learned by the video representation network to the image representation network. Specifically, given the input videos, we enforce the image representation network to fit the outputs of video representation network in a shared feature space. With back propagation, temporal knowledge can be transferred to enhance the image features and the information asymmetry problem can be alleviated. With additional classification and integrated triplet losses, our model can learn expressive and discriminative image and video features for image-to-video re-identification. Extensive experiments demonstrate the effectiveness of our method and the overall results on two widely used datasets surpass the state-of-the-art methods by a large margin.
Person re-identification (reID) benefits greatly from deep convolutional neural networks (CNNs) which learn robust feature embeddings. However, CNNs are inherently limited in modeling the large variations in person pose and scale due to their fixed geometric structures. In this paper, we propose a novel network structure, Interaction-and-Aggregation (IA), to enhance the feature representation capability of CNNs. Firstly, Spatial IA (SIA) module is introduced. It models the interdependencies between spatial features and then aggregates the correlated features corresponding to the same body parts. Unlike CNNs which extract features from fixed rectangle regions, SIA can adaptively determine the receptive fields according to the input person pose and scale. Secondly, we introduce Channel IA (CIA) module which selectively aggregates channel features to enhance the feature representation, especially for smallscale visual cues. Further, IA network can be constructed by inserting IA blocks into CNNs at any depth. We validate the effectiveness of our model for person reID by demonstrating its superiority over state-of-the-art methods on three benchmark datasets.
Video person re-identification (re-ID) plays an important role in surveillance video analysis. However, the performance of video re-ID degenerates severely under partial occlusion. In this paper, we propose a novel network, called Spatio-Temporal Completion network (STCnet), to explicitly handle partial occlusion problem. Different from most previous works that discard the occluded frames, STCnet can recover the appearance of the occluded parts. For one thing, the spatial structure of a pedestrian frame can be used to predict the occluded body parts from the unoccluded body parts of this frame. For another, the temporal patterns of pedestrian sequence provide important clues to generate the contents of occluded parts. With the Spatio-temporal information, STCnet can recover the appearance for the occluded parts, which could be leveraged with those unoccluded parts for more accurate video re-ID. By combining a re-ID network with STCnet, a video re-ID framework robust to partial occlusion (VRSTC) is proposed. Experiments on three challenging video re-ID databases demonstrate that the proposed approach outperforms the state-of-the-art.
Recent researches attempt to improve the detection performance by adopting the idea of cascade for single-stage detectors. In this paper, we analyze and discover that inconsistency is the major factor limiting the performance. The refined anchors are associated with the feature extracted from the previous location and the classifier is confused by misaligned classification and localization. Further, we point out two main designing rules for the cascade manner: improving consistency between classification confidence and localization performance, and maintaining feature consistency between different stages. A multistage object detector named Cas-RetinaNet, is then proposed for reducing the misalignments. It consists of sequential stages trained with increasing IoU thresholds for improving the correlation, and a novel Feature Consistency Module for mitigating the feature inconsistency. Experiments show that our proposed Cas-RetinaNet achieves stable performance gains across different models and input scales. Specifically, our method improves RetinaNet from 39.1 AP to 41.1 AP on the challenging MS COCO dataset without any bells or whistles.
Multi-view facial expression recognition (FER) is a challenging task because the appearance of an expression varies in poses. To alleviate the influences of poses, recent methods either perform pose normalization or learn separate FER classifiers for each pose. However, these methods usually have two stages and rely on good performance of pose estimators. Different from existing methods, we propose a pose-adaptive hierarchical attention network (PhaNet) that can jointly recognize the facial expressions and poses in unconstrained environment. Specifically, PhaNet discovers the most relevant regions to the facial expression by an attention mechanism in hierarchical scales, and the most informative scales are then selected to learn the pose-invariant and expression-discriminative representations. PhaNet is end-to-end trainable by minimizing the hierarchical attention losses, the FER loss and pose loss with dynamically learned loss weights. We validate the effectiveness of the proposed PhaNet on three multi-view datasets (BU-3DFE, Multi-pie, and KDEF) and two in-the-wild FER datasets (AffectNet and SFEW). Extensive experiments demonstrate that our framework outperforms the state-of-the-arts under both within-dataset and cross-dataset settings, achieving the average accuracies of 84.92\%, 93.53\%, 88.5\%, 54.82\% and 31.25\% respectively.
Weakly supervised object detection aims at learning precise object detectors, given image category labels. In recent prevailing works, this problem is generally formulated as a multiple instance learning module guided by an image classification loss. The object bounding box is assumed to be the one contributing most to the classification among all proposals. However, the region contributing most is also likely to be a crucial part or the supporting context of an object. To obtain a more accurate detector, in this work we propose a novel end-to-end weakly supervised detection approach, where a newly introduced generative adversarial segmentation module interacts with the conventional detection module in a collaborative loop. The collaboration mechanism takes full advantages of the complementary interpretations of the weakly supervised localization task, namely detection and segmentation tasks, forming a more comprehensive solution. Consequently, our method obtains more precise object bounding boxes, rather than parts or irrelevant surroundings. Expectedly, the proposed method achieves an accuracy of 51.0% on the PASCAL VOC 2007 dataset, outperforming the state-of-the-arts and demonstrating its superiority for weakly supervised object detection.
Benefitted from its great success on many tasks, deep learning is increasingly used on low-computational-cost devices, e.g. smartphone, embedded devices, etc. To reduce the high computational and memory cost, in this work, we propose a fully learnable group convolution module (FLGC for short) which is quite efficient and can be embedded into any deep neural networks for acceleration. Specifically, our proposed method automatically learns the group structure in the training stage in a fully end-to-end manner, leading to a better structure than the existing pre-defined, two-steps, or iterative strategies. Moreover, our method can be further combined with depthwise separable convolution, resulting in 5 times acceleration than the vanilla Resnet50 on single CPU. An additional advantage is that in our FLGC the number of groups can be set as any value, but not necessarily 2^k as in most existing methods, meaning better tradeoff between accuracy and speed. As evaluated in our experiments, our method achieves better performance than existing learnable group convolution and standard group convolution when using the same number of groups.