In this paper, we propose a method, named EqCo (Equivalent Rules for Contrastive Learning), to make self-supervised learning irrelevant to the number of negative samples in the contrastive learning framework. Inspired by the infomax principle, we point that the margin term in contrastive loss needs to be adaptively scaled according to the number of negative pairs in order to keep steady mutual information bound and gradient magnitude. EqCo bridges the performance gap among a wide range of negative sample sizes, so that for the first time, we can perform self-supervised contrastive training using only a few negative pairs (e.g.smaller than 256 per query) on large-scale vision tasks like ImageNet, while with little accuracy drop. This is quite a contrast to the widely used large batch training or memory bank mechanism in current practices. Equipped with EqCo, our simplified MoCo (SiMo) achieves comparable accuracy with MoCo v2 on ImageNet (linear evaluation protocol) while only involves 16 negative pairs per query instead of 65536, suggesting that large quantities of negative samples might not be a critical factor in contrastive learning frameworks.
Modern activation layers use non-linear functions to activate the neurons. In this paper, we present a simple but effective activation function we term ACON which learns to activate the neurons or not. Surprisingly, we find Swish, the recent popular NAS-searched activation, can be interpreted as a smooth approximation to ReLU. Intuitively, in the same way, we approximate the variants in the ReLU family to the Swish family, we call ACON, which makes Swish a special case of ACON and remarkably improves the performance. Next, we present meta-ACON, which explicitly learns to optimize the parameter switching between non-linear (activate) and linear (inactivate) and provides a new design space. By simply changing the activation function, we improve the ImageNet top-1 accuracy rate by 6.7% and 1.8% on MobileNet-0.25 and ResNet-152, respectively.
Question Answering (QA) over Knowledge Base (KB) aims to automatically answer natural language questions via well-structured relation information between entities stored in knowledge bases. In order to make KBQA more applicable in actual scenarios, researchers have shifted their attention from simple questions to complex questions, which require more KB triples and constraint inference. In this paper, we introduce the recent advances in complex QA. Besides traditional methods relying on templates and rules, the research is categorized into a taxonomy that contains two main branches, namely Information Retrieval-based and Neural Semantic Parsing-based. After describing the methods of these branches, we analyze directions for future research and introduce the models proposed by the Alime team.
We present a conceptually simple, flexible and effective framework for weight generating networks. Our approach is general that unifies two current distinct and extremely effective SENet and CondConv into the same framework on weight space. The method, called WeightNet, generalizes the two methods by simply adding one more grouped fully-connected layer to the attention activation layer. We use the WeightNet, composed entirely of (grouped) fully-connected layers, to directly output the convolutional weight. WeightNet is easy and memory-conserving to train, on the kernel space instead of the feature space. Because of the flexibility, our method outperforms existing approaches on both ImageNet and COCO detection tasks, achieving better Accuracy-FLOPs and Accuracy-Parameter trade-offs. The framework on the flexible weight space has the potential to further improve the performance. Code is available at https://github.com/megvii-model/WeightNet.
We present a conceptually simple but effective funnel activation for image recognition tasks, called Funnel activation (FReLU), that extends ReLU and PReLU to a 2D activation by adding a negligible overhead of spatial condition. The forms of ReLU and PReLU are y = max(x, 0) and y = max(x, px), respectively, while FReLU is in the form of y = max(x,T(x)), where T(x) is the 2D spatial condition. Moreover, the spatial condition achieves a pixel-wise modeling capacity in a simple way, capturing complicated visual layouts with regular convolutions. We conduct experiments on ImageNet, COCO detection, and semantic segmentation tasks, showing great improvements and robustness of FReLU in the visual recognition tasks. Code is available at https://github.com/megvii-model/FunnelAct.
Dense object detectors rely on the sliding-window paradigm that predicts the object over a regular grid of image. Meanwhile, the feature maps on the point of the grid are adopted to generate the bounding box predictions. The point feature is convenient to use but may lack the explicit border information for accurate localization. In this paper, We propose a simple and efficient operator called Border-Align to extract "border features" from the extreme point of the border to enhance the point feature. Based on the BorderAlign, we design a novel detection architecture called BorderDet, which explicitly exploits the border information for stronger classification and more accurate localization. With ResNet-50 backbone, our method improves single-stage detector FCOS by 2.8 AP gains (38.6 v.s. 41.4). With the ResNeXt-101-DCN backbone, our BorderDet obtains 50.3 AP, outperforming the existing state-of-the-art approaches. The code is available at (https://github.com/Megvii-BaseDetection/BorderDet).
In this paper, we propose an anchor-free object detector with a fully differentiable label assignment strategy, named AutoAssign. It automatically determines positive/negative samples by generating positive and negative weight maps to modify each location's prediction dynamically. Specifically, we present a center weighting module to adjust the category-specific prior distributions and a confidence weighting module to adapt the specific assign strategy of each instance. The entire label assignment process is differentiable and requires no additional modification to transfer to different datasets and tasks. Extensive experiments on MS COCO show that our method steadily surpasses other best sampling strategies by $ \sim $ 1\% AP with various backbones. Moreover, our best model achieves 52.1\% AP, outperforming all existing one-stage detectors. Besides, experiments on other datasets, \emph{e.g.}, PASCAL VOC, Objects365, and WiderFace, demonstrate the broad applicability of AutoAssign.
In this paper we propose a new intermediate supervision method, named LabelEnc, to boost the training of object detection systems. The key idea is to introduce a novel label encoding function, mapping the ground-truth labels into latent embedding, acting as an auxiliary intermediate supervision to the detection backbone during training. Our approach mainly involves a two-step training procedure. First, we optimize the label encoding function via an AutoEncoder defined in the label space, approximating the "desired" intermediate representations for the target object detector. Second, taking advantage of the learned label encoding function, we introduce a new auxiliary loss attached to the detection backbones, thus benefiting the performance of the derived detector. Experiments show our method improves a variety of detection systems by around 2% on COCO dataset, no matter one-stage or two-stage frameworks. Moreover, the auxiliary structures only exist during training, i.e. it is completely cost-free in inference time.