Semantic retrieval, which retrieves semantically matched items given a textual query, has been an essential component to enhance system effectiveness in e-commerce search. In this paper, we study the multimodal retrieval problem, where the visual information (e.g, image) of item is leveraged as supplementary of textual information to enrich item representation and further improve retrieval performance. Though learning from cross-modality data has been studied extensively in tasks such as visual question answering or media summarization, multimodal retrieval remains a non-trivial and unsolved problem especially in the asymmetric scenario where the query is unimodal while the item is multimodal. In this paper, we propose a novel model named SMAR, which stands for Semantic-enhanced Modality-Asymmetric Retrieval, to tackle the problem of modality fusion and alignment in this kind of asymmetric scenario. Extensive experimental results on an industrial dataset show that the proposed model outperforms baseline models significantly in retrieval accuracy. We have open sourced our industrial dataset for the sake of reproducibility and future research works.

Product images are essential for providing desirable user experience in an e-commerce platform. For a platform with billions of products, it is extremely time-costly and labor-expensive to manually pick and organize qualified images. Furthermore, there are the numerous and complicated image rules that a product image needs to comply in order to be generated/selected. To address these challenges, in this paper, we present a new learning framework in order to achieve Automatic Generation of Product-Image Sequence (AGPIS) in e-commerce. To this end, we propose a Multi-modality Unified Image-sequence Classifier (MUIsC), which is able to simultaneously detect all categories of rule violations through learning. MUIsC leverages textual review feedback as the additional training target and utilizes product textual description to provide extra semantic information. %Without using prior knowledge or manually-crafted task, a single MUIsC model is able to learn the holistic knowledge of image reviewing and detect all categories of rule violations simultaneously. Based on offline evaluations, we show that the proposed MUIsC significantly outperforms various baselines. Besides MUIsC, we also integrate some other important modules in the proposed framework, such as primary image selection, non-compliant content detection, and image deduplication. With all these modules, our framework works effectively and efficiently in JD.com recommendation platform. By Dec 2021, our AGPIS framework has generated high-standard images for about 1.5 million products and achieves 13.6% in reject rate. Code of this work is available at https://github.com/efan3000/muisc.

We propose a novel domain-specific generative pre-training (DSGPT) method for text generation and apply it to the product title and review summarization problems on E-commerce mobile display. First, we adopt a decoder-only transformer architecture, which fits well for fine-tuning tasks by combining input and output all together. Second, we demonstrate utilizing only small amount of pre-training data in related domains is powerful. Pre-training a language model from a general corpus such as Wikipedia or the Common Crawl requires tremendous time and resource commitment, and can be wasteful if the downstream tasks are limited in variety. Our DSGPT is pre-trained on a limited dataset, the Chinese short text summarization dataset (LCSTS). Third, our model does not require product-related human-labeled data. For title summarization task, the state of art explicitly uses additional background knowledge in training and predicting stages. In contrast, our model implicitly captures this knowledge and achieves significant improvement over other methods, after fine-tuning on the public Taobao.com dataset. For review summarization task, we utilize JD.com in-house dataset, and observe similar improvement over standard machine translation methods which lack the flexibility of fine-tuning. Our proposed work can be simply extended to other domains for a wide range of text generation tasks.

Human visual perception shows good consistency for many multi-label image classification tasks under certain spatial transforms, such as scaling, rotation, flipping and translation. This has motivated the data augmentation strategy widely used in CNN classifier training -- transformed images are included for training by assuming the same class labels as their original images. In this paper, we further propose the assumption of perceptual consistency of visual attention regions for classification under such transforms, i.e., the attention region for a classification follows the same transform if the input image is spatially transformed. While the attention regions of CNN classifiers can be derived as an attention heatmap in middle layers of the network, we find that their consistency under many transforms are not preserved. To address this problem, we propose a two-branch network with an original image and its transformed image as inputs and introduce a new attention consistency loss that measures the attention heatmap consistency between two branches. This new loss is then combined with multi-label image classification loss for network training. Experiments on three datasets verify the superiority of the proposed network by achieving new state-of-the-art classification performance.

In this paper, we study the problem of Cross-View Person Identification (CVPI), which aims at identifying the same person from temporally synchronized videos taken by different wearable cameras. Our basic idea is to utilize the human motion consistency for CVPI, where human motion can be computed by optical flow. However, optical flow is view-variant - the same person's optical flow in different videos can be very different due to view angle change. In this paper, we attempt to utilize 3D human-skeleton sequences to learn a model that can extract view-invariant motion features from optical flows in different views. For this purpose, we use 3D Mocap database to build a synthetic optical flow dataset and train a Triplet Network (TN) consisting of three sub-networks: two for optical flow sequences from different views and one for the underlying 3D Mocap skeleton sequence. Finally, sub-networks for optical flows are used to extract view-invariant features for CVPI. Experimental results show that, using only the motion information, the proposed method can achieve comparable performance with the state-of-the-art methods. Further combination of the proposed method with an appearance-based method achieves new state-of-the-art performance.

Automatic tracking of large-scale crowded targets are of particular importance in many applications, such as crowded people/vehicle tracking in video surveillance, fiber tracking in materials science, and cell tracking in biomedical imaging. This problem becomes very challenging when the targets show similar appearance and the interslice/ inter-frame continuity is low due to sparse sampling, camera motion and target occlusion. The main challenge comes from the step of association which aims at matching the predictions and the observations of the multiple targets. In this paper we propose a new groupwise method to explore the target group information and employ the within-group correlations for association and tracking. In particular, the within-group association is modeled by a nonrigid 2D Thin-Plate transform and a sequence of group shrinking, group growing and group merging operations are then developed to refine the composition of each group. We apply the proposed method to track large-scale fibers from microscopy material images and compare its performance against several other multi-target tracking methods. We also apply the proposed method to track crowded people from videos with poor inter-frame continuity.

We propose a new learning-based method for estimating 2D human pose from a single image, using Dual-Source Deep Convolutional Neural Networks (DS-CNN). Recently, many methods have been developed to estimate human pose by using pose priors that are estimated from physiologically inspired graphical models or learned from a holistic perspective. In this paper, we propose to integrate both the local (body) part appearance and the holistic view of each local part for more accurate human pose estimation. Specifically, the proposed DS-CNN takes a set of image patches (category-independent object proposals for training and multi-scale sliding windows for testing) as the input and then learns the appearance of each local part by considering their holistic views in the full body. Using DS-CNN, we achieve both joint detection, which determines whether an image patch contains a body joint, and joint localization, which finds the exact location of the joint in the image patch. Finally, we develop an algorithm to combine these joint detection/localization results from all the image patches for estimating the human pose. The experimental results show the effectiveness of the proposed method by comparing to the state-of-the-art human-pose estimation methods based on pose priors that are estimated from physiologically inspired graphical models or learned from a holistic perspective.

Wearable cameras, such as Google Glass and Go Pro, enable video data collection over larger areas and from different views. In this paper, we tackle a new problem of locating the co-interest person (CIP), i.e., the one who draws attention from most camera wearers, from temporally synchronized videos taken by multiple wearable cameras. Our basic idea is to exploit the motion patterns of people and use them to correlate the persons across different videos, instead of performing appearance-based matching as in traditional video co-segmentation/localization. This way, we can identify CIP even if a group of people with similar appearance are present in the view. More specifically, we detect a set of persons on each frame as the candidates of the CIP and then build a Conditional Random Field (CRF) model to select the one with consistent motion patterns in different videos and high spacial-temporal consistency in each video. We collect three sets of wearable-camera videos for testing the proposed algorithm. All the involved people have similar appearances in the collected videos and the experiments demonstrate the effectiveness of the proposed algorithm.