Our work focuses on tackling the challenging but natural visual recognition task of long-tailed data distribution (i.e., a few classes occupy most data, while have rarely samples). In literature, class re-balancing strategies (e.g., re-weighting and re-sampling) are prominent effective methods proposed to alleviate extreme imbalance for dealing with problems. this paper, we firstly discover that these achieving satisfactory accuracy owe they could significantly promote classifier learning...

State-of-the-art distillation methods are mainly based on distilling deep features from intermediate layers, while the significance of logit is greatly overlooked. To provide a novel viewpoint to study distillation, we re-formulate classical KD loss into two parts, i.e., target class knowledge (TCKD) and non-target (NCKD). We empirically investigate prove effects parts: TCKD transfers concerning "difficulty" training samples, NCKD prominent reason why works. More importantly, reveal that...

Over recent years, emerging interest has occurred in integrating computer vision technology into the retail industry. Automatic checkout (ACO) is one of critical problems this area which aims to automatically generate shopping list from images products purchase. The main challenge problem comes large scale and fine-grained nature product categories as well difficulty for collecting training that reflect realistic scenarios due continuous update products. Despite its significant practical...

Computer vision (CV) is the process of using machines to understand and analyze imagery, which an integral branch artificial intelligence. Among various research areas CV, fine-grained image analysis (FGIA) a longstanding fundamental problem, has become ubiquitous in diverse real-world applications. The task FGIA targets analyzing visual objects from subordinate categories, \eg, species birds or models cars. small inter-class variations large intra-class caused by nature makes it challenging...

Multi-label image recognition has attracted considerable research attention and achieved great success in recent years. Capturing label correlations is an effective manner to advance the performance of multi-label recognition. Two types were principally studied, i.e., spatial semantic correlations. However, literature, previous methods considered only either them. In this work, inspired by Transformer, we propose a plug-and-play module, named Spatial Semantic Transformers (SST),...

Text-supervised semantic segmentation is a novel research topic that allows segments to emerge with image-text contrasting. However, pioneering methods could be subject specifically designed network architectures. This paper shows vanilla contrastive language-image pretraining (CLIP) model an effective text-supervised segmentor by itself. First, we reveal CLIP inferior localization and due its optimization being driven densely aligning visual language representations. Second, propose the...

Our work focuses on tackling the challenging but natural visual recognition task of long-tailed data distribution (i.e., a few classes occupy most data, while have rarely samples). In literature, class re-balancing strategies (e.g., re-weighting and re-sampling) are prominent effective methods proposed to alleviate extreme imbalance for dealing with problems. this paper, we firstly discover that these achieving satisfactory accuracy owe they could significantly promote classifier learning...

Multi-label image recognition is a fundamental but challenging computer vision and multimedia task. Great progress has been achieved by exploiting label correlations among these multiple labels associated with single image, which the most crucial issue for multi-label recognition. In this paper, to explicitly model correlations, we propose unified deep learning framework Disentangle, Embed Rank (DER) corresponding cues. Specifically, first obtain class-aware disentangled maps (CADMs)...

Under the background of increasing prosperity Internet finance, quantitative investment has become a hot topic, among which prediction stock price is focus research.In this paper, an optimized nonlinear integration framework based on feature clustering and deep learning proposed to predict daily data.Clustering algorithm used divide complex changeable data into multiple clusters according its characteristics, can pave way for establishment forecast model.Bidirectional long short-term memory...

Searching for the nearest neighbor is a fundamental problem in computer vision field, and deep hashing has become one of most representative widely used methods, which learns to generate compact binary codes visual data. In this paper, we first delve into representation learning surprisingly find that could be double-edged sword, i.e., can accelerate query speed decrease storage cost search progress, but it greatly sacrifices discriminability representations especially with extremely short...

Multi-label image recognition with convolutional neural networks has achieved remarkable progress in the past few years. However, most existing multi-label methods suffer from long-tailed data distribution problem, \ie, head categories occupy training samples, while tailed classes have samples. This work firstly studies influence of on methods. Based this, two crucial issues are identified: 1) severe gradient imbalance between and categories, even though re-balancing strategies adopted; 2)...

Retrieving content relevant images from a large-scale fine-grained dataset could suffer intolerably slow query speed and highly redundant storage cost, due to high-dimensional real-valued embeddings which aim distinguish subtle visual differences of objects. In this paper, we study the novel hashing topic generate compact binary codes for images, leveraging search efficiency hash learning alleviate aforementioned problems. Specifically, propose unified end-to-end trainable network, termed as...