DETR是一种新颖的端到端transformer架构目标检测器，在规模扩大时显著优于经典检测器。本文专注于通过知识蒸馏压缩DETR。尽管知识蒸馏在经典检测器中已被充分研究，但缺乏如何使其在DETR上有效工作的研究。我们首先提供实验和理论分析，指出DETR蒸馏的主要挑战在于缺乏一致的蒸馏点。蒸馏点指供学生模型模仿的预测对应输入，其在CNN检测器和DETR中有不同表述，可靠蒸馏需要教师与学生间足够多且一致的蒸馏点。基于此观察，我们提出首个面向DETR的通用知识蒸馏范式（KD-DETR），采用一致蒸馏点采样，适用于同构和异构蒸馏。具体而言，我们通过引入一组专用object queries为DETR构建蒸馏点，解耦检测与蒸馏任务。我们进一步提出由通用到特定的蒸馏点采样策略，探索KD-DETR的可扩展性。大量实验验证了KD-DETR的有效性与泛化性。对于单尺度DAB-DETR及多尺度Deformable DETR和DINO，KD-DETR使学生模型性能提升2.6%-5.2%。我们进一步将KD-DETR扩展至异构蒸馏，通过将DINO知识蒸馏至Faster R-CNN with ResNet-50实现2.1%提升，与同构蒸馏方法相当。代码见https://github.com/wennyuhey/KD-DETR。
英文摘要：
DETR is a novel end-to-end transformer architecture object detector, which significantly outperforms classic detectors when scaling up. In this paper, we focus on the compression of DETR with knowledge distillation. While knowledge distillation has been well-studied in classic detectors, there is a lack of researches on how to make it work effectively on DETR. We first provide experimental and theoretical analysis to point out that the main challenge in DETR distillation is the lack of consistent distillation points. Distillation points refer to the corresponding inputs of the predictions for student to mimic, which have different formulations in CNN detector and DETR, and reliable distillation requires sufficient distillation points which are consistent between teacher and student.
Based on this observation, we propose the first general knowledge distillation paradigm for DETR (KD-DETR) with consistent distillation points sampling, for both homogeneous and heterogeneous distillation. Specifically, we decouple detection and distillation tasks by introducing a set of specialized object queries to construct distillation points for DETR. We further propose a general-to-specific distillation points sampling strategy to explore the extensibility of KD-DETR. Extensive experiments validate the effectiveness and generalization of KD-DETR. For both single-scale DAB-DETR and multis-scale Deformable DETR and DINO, KD-DETR boost the performance of student model with improvements of $2.6\%-5.2\%$. We further extend KD-DETR to heterogeneous distillation, and achieves $2.1\%$ improvement by distilling the knowledge from DINO to Faster R-CNN with ResNet-50, which is comparable with homogeneous distillation methods.The code is available at https://github.com/wennyuhey/KD-DETR.
University of Science and Technology of China（中国科学技术大学）；University of Edinburgh（爱丁堡大学）；University of Adelaide（阿德莱德大学）
Transformer-based detectors (DETRs) are becoming popular for their simple framework, but the large model size and heavy time consumption hinder their deployment in the real world. While knowledge distillation (KD) can be an appealing technique to compress giant detectors into small ones for comparable detection performance and low inference cost. Since DETRs formulate object detection as a set prediction problem, existing KD methods designed for classic convolution-based detectors may not be directly applicable. In this paper, we propose DETRDistill, a novel knowledge distillation method dedic…
The recent detection transformer (DETR) simplifies the object detection pipeline by removing hand-crafted designs and hyperparameters as employed in conventional two-stage object detectors. However, how to leverage the simple yet effective DETR architecture in domain adaptive object detection is largely neglected. Inspired by the unique DETR attention mechanisms, we design DA-DETR, a domain adaptive object detection transformer that introduces information fusion for effective transfer from a labeled source domain to an unlabeled target domain. DA-DETR introduces a novel CNN-Transformer Blender…
Peking University（北京大学）；Beihang University（北京航空航天大学）；Microsoft Research Asia（微软亚洲研究院）
While various knowledge distillation (KD) methods in CNN-based detectors show their effectiveness in improving small students, the baselines and recipes for DETR-based detectors are yet to be built. In this paper, we focus on the transformer decoder of DETR-based detectors and explore KD methods for them. The outputs of the transformer decoder lie in random order, which gives no direct correspondence between the predictions of the teacher and the student, thus posing a challenge for knowledge distillation. To this end, we propose MixMatcher to align the decoder outputs of DETR-based teachers a…
National Yang Ming Chiao Tung University（国立阳明交通大学）
Recently, Transformer-based methods have been utilized to improve the performance of human action recognition. However, most of these studies assume that multi-view data is complete, which may not always be the case in real-world scenarios. Therefore, this paper presents a novel Multi-view Knowledge Distillation Transformer (MKDT) framework that consists of a teacher network and a student network. This framework aims to handle incomplete human action problems in real-world applications. Specifically, the multi-view knowledge distillation transformer uses a hierarchical vision transformer with …
Recent self-supervised methods are mainly designed for representation learning with the base model, e.g., ResNets or ViTs. They cannot be easily transferred to DETR, with task-specific Transformer modules. In this work, we present Siamese DETR, a Siamese self-supervised pretraining approach for the Transformer architecture in DETR. We consider learning view-invariant and detection-oriented representations simultaneously through two complementary tasks, i.e., localization and discrimination, in a novel multi-view learning framework. Two self-supervised pretext tasks are designed: (i) Multi-View…
Recently, neural networks purely based on attention were shown to address image understanding tasks such as image classification. However, these visual transformers are pre-trained with hundreds of millions of images using an expensive infrastructure, thereby limiting their adoption. In this work, we produce a competitive convolution-free transformer by training on Imagenet only. We train them on a single computer in less than 3 days. Our reference vision transformer (86M parameters) achieves top-1 accuracy of 83.1% (single-crop evaluation) on ImageNet with no external data. More importantly, …
ShanghaiTech University（上海科技大学）；JD Explore Academy（京东探索研究院）；Meituan Inc.（美团公司）；The University of Sydney（悉尼大学）；Shanghai Engineering Research Center of Intelligent Vision and Imaging（上海智能视觉与成像工程技术研究中心）；Shanghai Engineering Research Center of Energy Efficient and Custom AI IC（上海能效与定制人工智能集成电路工程技术研究中心）
Transformers are successfully applied to computer vision due to their powerful modeling capacity with self-attention. However, the excellent performance of transformers heavily depends on enormous training images. Thus, a data-efficient transformer solution is urgently needed. In this work, we propose an early knowledge distillation framework, which is termed as DearKD, to improve the data efficiency required by transformers. Our DearKD is a two-stage framework that first distills the inductive biases from the early intermediate layers of a CNN and then gives the transformer full play by train…
Computer Vision Center（计算机视觉中心）；Universitat Autònoma de Barcelona（巴塞罗那自治大学）
Self-supervised learning has been widely applied to train high-quality vision transformers. Unleashing their excellent performance on memory and compute constraint devices is therefore an important research topic. However, how to distill knowledge from one self-supervised ViT to another has not yet been explored. Moreover, the existing self-supervised knowledge distillation (SSKD) methods focus on ConvNet based architectures are suboptimal for ViT knowledge distillation. In this paper, we study knowledge distillation of self-supervised vision transformers (ViT-SSKD). We show that directly dist…
Beijing Institute of Technology（北京理工大学）；Huawei Technologies（华为技术有限公司）；Peking University（北京大学）
In the past few years, transformers have achieved promising performances on various computer vision tasks. Unfortunately, the immense inference overhead of most existing vision transformers withholds their from being deployed on edge devices such as cell phones and smart watches. Knowledge distillation is a widely used paradigm for compressing cumbersome architectures via transferring information to a compact student. However, most of them are designed for convolutional neural networks (CNNs), which do not fully investigate the character of vision transformer (ViT). In this paper, we utilize t…
Scaled dot-product attention applies a softmax function on the scaled dot-product of queries and keys to calculate weights and then multiplies the weights and values. In this work, we study how to improve the learning of scaled dot-product attention to improve the accuracy of DETR. Our method is based on the following observations: using ground truth foreground-background mask (GT Fg-Bg Mask) as additional cues in the weights/values learning enables learning much better weights/values; with better weights/values, better values/weights can be learned. We propose a triple-attention module in whi…
Microsoft Research（微软研究院）；Microsoft Cloud+AI（微软云与人工智能部门）
Vision transformer (ViT) recently has drawn great attention in computer vision due to its remarkable model capability. However, most prevailing ViT models suffer from huge number of parameters, restricting their applicability on devices with limited resources. To alleviate this issue, we propose TinyViT, a new family of tiny and efficient small vision transformers pretrained on large-scale datasets with our proposed fast distillation framework. The central idea is to transfer knowledge from large pretrained models to small ones, while enabling small models to get the dividends of massive pretr…
South China University of Technology（华南理工大学）；MIT（麻省理工学院）；University of Texas at Austin（得克萨斯大学奥斯汀分校）；Tsinghua University（清华大学）；Shanghai Qi Zhi Institute（上海期智研究院）
Transformers recently are adapted from the community of natural language processing as a promising substitute of convolution-based neural networks for visual learning tasks. However, its supremacy degenerates given an insufficient amount of training data (e.g., ImageNet). To make it into practical utility, we propose a novel distillation-based method to train vision transformers. Unlike previous works, where merely heavy convolution-based teachers are provided, we introduce lightweight teachers with different architectural inductive biases (e.g., convolution and involution) to co-advise the st…
LIB-KD: Teaching Inductive Bias for Efficient Vision Transformer Distillation and Compression
Department of Electrical Engineering Indian Institute of Technology Delhi（电子工程系印度理工学院德里）；Yogananda School of AI, Computers and Data Sciences（Yoganananda人工智能、计算机与数据科学学院）
Weight Copy and Low-Rank Adaptation for Few-Shot Distillation of Vision Transformers
University of Bucharest（布加勒斯特大学）；Norwegian University of Science and Technology（挪威科技大学）；University of Medicine and Pharmacy “Carol Davila”（卡罗尔·达维拉医药大学）
University of Science and Technology of China（中国科学技术大学）；SenseTime Research（商汤科技研究院）；The Chinese University of Hong Kong（香港中文大学）
DETR has been recently proposed to eliminate the need for many hand-designed components in object detection while demonstrating good performance. However, it suffers from slow convergence and limited feature spatial resolution, due to the limitation of Transformer attention modules in processing image feature maps. To mitigate these issues, we proposed Deformable DETR, whose attention modules only attend to a small set of key sampling points around a reference. Deformable DETR can achieve better performance than DETR (especially on small objects) with 10 times less training epochs. Extensive e…
Rekor Systems, Inc.（Rekor Systems 公司）；University of Kansas（堪萨斯大学）；Ryerson University（瑞尔森大学）
Object Detection with Transformers (DETR) and related works reach or even surpass the highly-optimized Faster-RCNN baseline with self-attention network architectures. Inspired by the evidence that pure self-attention possesses a strong inductive bias that leads to the transformer losing the expressive power with respect to network depth, we propose a transformer architecture with a mitigatory self-attention mechanism by applying possible direct mapping connections in the transformer architecture to mitigate the rank collapse so as to counteract feature expression loss and enhance the model per…
Peking University（北京大学）；Microsoft Research Asia（微软亚洲研究院）
In this paper, we are interested in Detection Transformer (DETR), an end-to-end object detection approach based on a transformer encoder-decoder architecture without hand-crafted postprocessing, such as NMS. Inspired by Conditional DETR, an improved DETR with fast training convergence, that presented box queries (originally called spatial queries) for internal decoder layers, we reformulate the object query into the format of the box query that is a composition of the embeddings of the reference point and the transformation of the box with respect to the reference point. This reformulation ind…
Institute of Computing Technology (ICT), Chinese Academy of Sciences（中国科学院计算技术研究所）；Stanford University（斯坦福大学）；University of Chinese Academy of Sciences（中国科学院大学）；AI Lab, Lenovo Research（联想研究院人工智能实验室）；Lenovo Ltd.（联想集团有限公司）
Recently, the dominant DETR-based approaches apply central-concept spatial prior to accelerate Transformer detector convergency. These methods gradually refine the reference points to the center of target objects and imbue object queries with the updated central reference information for spatially conditional attention. However, centralizing reference points may severely deteriorate queries' saliency and confuse detectors due to the indiscriminative spatial prior. To bridge the gap between the reference points of salient queries and Transformer detectors, we propose SAlient Point-based DETR (S…
International Digital Economy Academy (IDEA)（国际数字经济研究院（IDEA））；University of Science and Technology of China（中国科学技术大学）；Peking University（北京大学）；Microsoft Research Asia（微软亚洲研究院）；Microsoft Research, Redmond（微软雷德蒙德研究院）
SenseTime Research（商汤科技研究院）；Institute of Automation, CAS（中国科学院自动化研究所）；Peng Cheng Lab（鹏城实验室）
Self-supervised pre-training and transformer-based networks have significantly improved the performance of object detection. However, most of the current self-supervised object detection methods are built on convolutional-based architectures. We believe that the transformers' sequence characteristics should be considered when designing a transformer-based self-supervised method for the object detection task. To this end, we propose SeqCo-DETR, a novel Sequence Consistency-based self-supervised method for object DEtection with TRansformers. SeqCo-DETR defines a simple but effective pretext by m…
Recent proposed DETR variants have made tremendous progress in various scenarios due to their streamlined processes and remarkable performance. However, the learned queries usually explore the global context to generate the final set prediction, resulting in redundant burdens and unfaithful results. More specifically, a query is commonly responsible for objects of different scales and positions, which is a challenge for the query itself, and will cause spatial resource competition among queries. To alleviate this issue, we propose Team DETR, which leverages query collaboration and position con…
We consider the problem of omni-supervised object detection, which can use unlabeled, fully labeled and weakly labeled annotations, such as image tags, counts, points, etc., for object detection. This is enabled by a unified architecture, Omni-DETR, based on the recent progress on student-teacher framework and end-to-end transformer based object detection. Under this unified architecture, different types of weak labels can be leveraged to generate accurate pseudo labels, by a bipartite matching based filtering mechanism, for the model to learn. In the experiments, Omni-DETR has achieved state-…
Recently, Transformer-based methods, which predict polygon points or Bezier curve control points for localizing texts, are popular in scene text detection. However, these methods built upon detection transformer framework might achieve sub-optimal training efficiency and performance due to coarse positional query modeling.In addition, the point label form exploited in previous works implies the reading order of humans, which impedes the detection robustness from our observation. To address these challenges, this paper proposes a concise Dynamic Point Text DEtection TRansformer network, termed …
National University of Singapore（新加坡国立大学）；Yitu Technology（依图科技）
Recently, DETR pioneered the solution of vision tasks with transformers, it directly translates the image feature map into the object detection result. Though effective, translating the full feature map can be costly due to redundant computation on some area like the background. In this work, we encapsulate the idea of reducing spatial redundancy into a novel poll and pool (PnP) sampling module, with which we build an end-to-end PnP-DETR architecture that adaptively allocates its computation spatially to be more efficient. Concretely, the PnP module abstracts the image feature map into fine fo…
Dual-R-DETR: Resolving Query Competition with Pairwise Routing in Transformer Decoders
University of Pittsburgh（匹兹堡大学）；University of California, Irvine（加州大学尔湾分校）；Stevens Institute of Technology（史蒂文斯理工学院）；Illinois Institute of Technology（伊利诺伊理工学院）；New York University（纽约大学）
