In a typical multi-label setting, a picture contains on average few positive labels, and many negative ones. This positive-negative imbalance dominates the optimization process, and can lead to under-emphasizing gradients from positive labels during training, resulting in poor accuracy. In this paper, we introduce a novel asymmetric loss ("ASL"), which operates differently on positive and negative samples. The loss enables to dynamically down-weights and hard-thresholds easy negative samples, while also discarding possibly mislabeled samples. We demonstrate how ASL can balance the probabilities of different samples, and how this balancing is translated to better mAP scores. With ASL, we reach state-of-the-art results on multiple popular multi-label datasets: MS-COCO, Pascal-VOC, NUS-WIDE and Open Images. We also demonstrate ASL applicability for other tasks, such as single-label classification and object detection. ASL is effective, easy to implement, and does not increase the training time or complexity. Implementation is available at: https://github.com/Alibaba-MIIL/ASL.

The task of person re-identification (ReID) has attracted growing attention in recent years leading to improved performance, albeit with little focus on real-world applications. Most SotA methods are based on heavy pre-trained models, e.g. ResNet50 (~25M parameters), which makes them less practical and more tedious to explore architecture modifications. In this study, we focus on a small-sized randomly initialized model that enables us to easily introduce architecture and training modifications suitable for person ReID. The outcomes of our study are a compact network and a fitting training regime. We show the robustness of the network by outperforming the SotA on both Market1501 and DukeMTMC. Furthermore, we show the representation power of our ReID network via SotA results on a different task of multi-object tracking.

Single image interpolation is a central and extensively studied problem in image processing. A common approach toward the treatment of this problem in recent years is to divide the given image into overlapping patches and process each of them based on a model for natural image patches. Adaptive sparse representation modeling is one such promising image prior, which has been shown to be powerful in filling-in missing pixels in an image. Another force that such algorithms may use is the self-similarity that exists within natural images. Processing groups of related patches together exploits their correspondence, leading often times to improved results. In this paper, we propose a novel image interpolation method, which combines these two forces-nonlocal self-similarities and sparse representation modeling. The proposed method is contrasted with competitive and related algorithms, and demonstrated to achieve state-of-the-art results.

In this paper, we consider denoising of image sequences that are corrupted by zero-mean additive white Gaussian noise. Relative to single image denoising techniques, denoising of sequences aims to also utilize the temporal dimension. This assists in getting both faster algorithms and better output quality. This paper focuses on utilizing sparse and redundant representations for image sequence denoising, extending the work reported in. In the single image setting, the K-SVD algorithm is used to train a sparsifying dictionary for the corrupted image. This paper generalizes the above algorithm by offering several extensions: i) the atoms used are 3-D; ii) the dictionary is propagated from one frame to the next, reducing the number of required iterations; and iii) averaging is done on patches in both spatial and temporal neighboring locations. These modifications lead to substantial benefits in complexity and denoising performance, compared to simply running the single image algorithm sequentially. The algorithm's performance is experimentally compared to several state-of-the-art algorithms, demonstrating comparable or favorable results.

Super-resolution reconstruction (SRR) has long been relying on very accurate motion estimation between the frames for a successful process. However, recent works propose SRR that bypasses the need for an explicit motion estimation. In this correspondence, we present a new framework that ultimately leads to the same algorithm as in our prior work. The contribution of this paper is two-fold. First, the suggested approach is much simpler and more intuitive, relying on the classic SRR formulation, and using a probabilistic and crude motion estimation. Second, the new approach offers various extensions not covered in our previous work, such as more general re-sampling tasks (e.g., de-interlacing).

Super-resolution reconstruction proposes a fusion of several low-quality images into one higher quality result with better optical resolution. Classic super-resolution techniques strongly rely on the availability of accurate motion estimation for this fusion task. When the motion is estimated inaccurately, as often happens for nonglobal motion fields, annoying artifacts appear in the super-resolved outcome. Encouraged by recent developments on the video denoising problem, where state-of-the-art algorithms are formed with no explicit motion estimation, we seek a super-resolution algorithm of similar nature that will allow processing sequences with general motion patterns. In this paper, we base our solution on the Nonlocal-Means (NLM) algorithm. We show how this denoising method is generalized to become a relatively simple super-resolution algorithm with no explicit motion estimation. Results on several test movies show that the proposed method is very successful in providing super-resolution on general sequences.

The need for precise (subpixel accuracy) motion estimates in conventional super-resolution has limited its applicability to only video sequences with relatively simple motions such as global translational or affine displacements. In this paper, we introduce a novel framework for adaptive enhancement and spatiotemporal upscaling of videos containing complex activities without explicit need for accurate motion estimation. Our approach is based on multidimensional kernel regression, where each pixel in the video sequence is approximated with a 3-D local (Taylor) series, capturing the essential local behavior of its spatiotemporal neighborhood. The coefficients of this series are estimated by solving a local weighted least-squares problem, where the weights are a function of the 3-D space-time orientation in the neighborhood. As this framework is fundamentally based upon the comparison of neighboring pixels in both space and time, it implicitly contains information about the local motion of the pixels across time, therefore rendering unnecessary an explicit computation of motions of modest size. The proposed approach not only significantly widens the applicability of super-resolution methods to a broad variety of video sequences containing complex motions, but also yields improved overall performance. Using several examples, we illustrate that the developed algorithm has super-resolution capabilities that provide improved optical resolution in the output, while being able to work on general input video with essentially arbitrary motion.