End-to-end paradigms significantly improve the accuracy of various deep-learning-based computer vision models. To this end, tasks like object detection have been upgraded by replacing non-end-to-end components, such as removing non-maximum suppression by training with a set loss based on bipartite matching. However, such an upgrade is not applicable to instance segmentation, due to its significantly higher output dimensions compared to object detection. In this paper, we propose an instance segmentation Transformer, termed ISTR, which is the first end-to-end framework of its kind. ISTR predicts low-dimensional mask embeddings, and matches them with ground truth mask embeddings for the set loss. Besides, ISTR concurrently conducts detection and segmentation with a recurrent refinement strategy, which provides a new way to achieve instance segmentation compared to the existing top-down and bottom-up frameworks. Benefiting from the proposed end-to-end mechanism, ISTR demonstrates state-of-the-art performance even with approximation-based suboptimal embeddings. Specifically, ISTR obtains a 46.8/38.6 box/mask AP using ResNet50-FPN, and a 48.1/39.9 box/mask AP using ResNet101-FPN, on the MS COCO dataset. Quantitative and qualitative results reveal the promising potential of ISTR as a solid baseline for instance-level recognition. Code has been made available at: https://github.com/hujiecpp/ISTR.
Model stealing attack aims to create a substitute model that steals the ability of the victim target model. However, most of the existing methods depend on the full probability outputs from the victim model, which is unavailable in most realistic scenarios. Focusing on the more practical hard-label setting, due to the lack of rich information in the probability prediction, the existing methods suffer from catastrophic performance degradation. Inspired by knowledge distillation, we propose a novel hard-label model stealing method termed \emph{black-box dissector}, which includes a CAM-driven erasing strategy to mine the hidden information in hard labels from the victim model, and a random-erasing-based self-knowledge distillation module utilizing soft labels from substitute model to avoid overfitting and miscalibration caused by hard labels. Extensive experiments on four widely-used datasets consistently show that our method outperforms state-of-the-art methods, with an improvement of at most $9.92\%$. In addition, experiments on real-world APIs further prove the effectiveness of our method. Our method also can invalidate existing defense methods which further demonstrates the practical potential of our methods.
Channel Pruning has been long adopted for compressing CNNs, which significantly reduces the overall computation. Prior works implement channel pruning in an unexplainable manner, which tends to reduce the final classification errors while failing to consider the internal influence of each channel. In this paper, we conduct channel pruning in a white box. Through deep visualization of feature maps activated by different channels, we observe that different channels have a varying contribution to different categories in image classification. Inspired by this, we choose to preserve channels contributing to most categories. Specifically, to model the contribution of each channel to differentiating categories, we develop a class-wise mask for each channel, implemented in a dynamic training manner w.r.t. the input image's category. On the basis of the learned class-wise mask, we perform a global voting mechanism to remove channels with less category discrimination. Lastly, a fine-tuning process is conducted to recover the performance of the pruned model. To our best knowledge, it is the first time that CNN interpretability theory is considered to guide channel pruning. Extensive experiments demonstrate the superiority of our White-Box over many state-of-the-arts. For instance, on CIFAR-10, it reduces 65.23% FLOPs with even 0.62% accuracy improvement for ResNet-110. On ILSVRC-2012, White-Box achieves a 45.6% FLOPs reduction with only a small loss of 0.83% in the top-1 accuracy for ResNet-50. Code, training logs and pruned models are anonymously at https://github.com/zyxxmu/White-Box.
While self-supervised representation learning (SSL) has received widespread attention from the community, recent research argue that its performance will suffer a cliff fall when the model size decreases. The current method mainly relies on contrastive learning to train the network and in this work, we propose a simple yet effective Distilled Contrastive Learning (DisCo) to ease the issue by a large margin. Specifically, we find the final embedding obtained by the mainstream SSL methods contains the most fruitful information, and propose to distill the final embedding to maximally transmit a teacher's knowledge to a lightweight model by constraining the last embedding of the student to be consistent with that of the teacher. In addition, in the experiment, we find that there exists a phenomenon termed Distilling BottleNeck and present to enlarge the embedding dimension to alleviate this problem. Our method does not introduce any extra parameter to lightweight models during deployment. Experimental results demonstrate that our method achieves the state-of-the-art on all lightweight models. Particularly, when ResNet-101/ResNet-50 is used as teacher to teach EfficientNet-B0, the linear result of EfficientNet-B0 on ImageNet is very close to ResNet-101/ResNet-50, but the number of parameters of EfficientNet-B0 is only 9.4%/16.3% of ResNet-101/ResNet-50.
Network pruning is an effective approach to reduce network complexity without performance compromise. Existing studies achieve the sparsity of neural networks via time-consuming weight tuning or complex search on networks with expanded width, which greatly limits the applications of network pruning. In this paper, we show that high-performing and sparse sub-networks without the involvement of weight tuning, termed "lottery jackpots", exist in pre-trained models with unexpanded width. For example, we obtain a lottery jackpot that has only 10% parameters and still reaches the performance of the original dense VGGNet-19 without any modifications on the pre-trained weights. Furthermore, we observe that the sparse masks derived from many existing pruning criteria have a high overlap with the searched mask of our lottery jackpot, among which, the magnitude-based pruning results in the most similar mask with ours. Based on this insight, we initialize our sparse mask using the magnitude pruning, resulting in at least 3x cost reduction on the lottery jackpot search while achieves comparable or even better performance. Specifically, our magnitude-based lottery jackpot removes 90% weights in the ResNet-50, while easily obtains more than 70% top-1 accuracy using only 10 searching epochs on ImageNet.
Few-shot class-incremental learning (FSCIL), which targets at continuously expanding model's representation capacity under few supervisions, is an important yet challenging problem. On the one hand, when fitting new tasks (novel classes), features trained on old tasks (old classes) could significantly drift, causing catastrophic forgetting. On the other hand, training the large amount of model parameters with few-shot novel-class examples leads to model over-fitting. In this paper, we propose a learnable expansion-and-compression network (LEC-Net), with the aim to simultaneously solve catastrophic forgetting and model over-fitting problems in a unified framework. By tentatively expanding network nodes, LEC-Net enlarges the representation capacity of features, alleviating feature drift of old network from the perspective of model regularization. By compressing the expanded network nodes, LEC-Net purses minimal increase of model parameters, alleviating over-fitting of the expanded network from a perspective of compact representation. Experiments on the CUB/CIFAR-100 datasets show that LEC-Net improves the baseline by 5~7% while outperforms the state-of-the-art by 5~6%. LEC-Net also demonstrates the potential to be a general incremental learning approach with dynamic model expansion capability.
Existing online knowledge distillation approaches either adopt the student with the best performance or construct an ensemble model for better holistic performance. However, the former strategy ignores other students' information, while the latter increases the computational complexity. In this paper, we propose a novel method for online knowledge distillation, termed FFSD, which comprises two key components: Feature Fusion and Self-Distillation, towards solving the above problems in a unified framework. Different from previous works, where all students are treated equally, the proposed FFSD splits them into a student leader and a common student set. Then, the feature fusion module converts the concatenation of feature maps from all common students into a fused feature map. The fused representation is used to assist the learning of the student leader. To enable the student leader to absorb more diverse information, we design an enhancement strategy to increase the diversity among students. Besides, a self-distillation module is adopted to convert the feature map of deeper layers into a shallower one. Then, the shallower layers are encouraged to mimic the transformed feature maps of the deeper layers, which helps the students to generalize better. After training, we simply adopt the student leader, which achieves superior performance, over the common students, without increasing the storage or inference cost. Extensive experiments on CIFAR-100 and ImageNet demonstrate the superiority of our FFSD over existing works. The code is available at https://github.com/SJLeo/FFSD.
Towards better unsupervised domain adaptation (UDA). Recently, researchers propose various domain-conditioned attention modules and make promising progresses. However, considering that the configuration of attention, i.e., the type and the position of attention module, affects the performance significantly, it is more generalized to optimize the attention configuration automatically to be specialized for arbitrary UDA scenario. For the first time, this paper proposes EvoADA: a novel framework to evolve the attention configuration for a given UDA task without human intervention. In particular, we propose a novel search space containing diverse attention configurations. Then, to evaluate the attention configurations and make search procedure UDA-oriented (transferability + discrimination), we apply a simple and effective evaluation strategy: 1) training the network weights on two domains with off-the-shelf domain adaptation methods; 2) evolving the attention configurations under the guide of the discriminative ability on the target domain. Experiments on various kinds of cross-domain benchmarks, i.e., Office-31, Office-Home, CUB-Paintings, and Duke-Market-1510, reveal that the proposed EvoADA consistently boosts multiple state-of-the-art domain adaptation approaches, and the optimal attention configurations help them achieve better performance.
Binary neural networks (BNNs) have received increasing attention due to their superior reductions of computation and memory. Most existing works focus on either lessening the quantization error by minimizing the gap between the full-precision weights and their binarization or designing a gradient approximation to mitigate the gradient mismatch, while leaving the "dead weights" untouched. This leads to slow convergence when training BNNs. In this paper, for the first time, we explore the influence of "dead weights" which refer to a group of weights that are barely updated during the training of BNNs, and then introduce rectified clamp unit (ReCU) to revive the "dead weights" for updating. We prove that reviving the "dead weights" by ReCU can result in a smaller quantization error. Besides, we also take into account the information entropy of the weights, and then mathematically analyze why the weight standardization can benefit BNNs. We demonstrate the inherent contradiction between minimizing the quantization error and maximizing the information entropy, and then propose an adaptive exponential scheduler to identify the range of the "dead weights". By considering the "dead weights", our method offers not only faster BNN training, but also state-of-the-art performance on CIFAR-10 and ImageNet, compared with recent methods. Code can be available at [this https URL](https://github.com/z-hXu/ReCU).
Few-shot object detection has made substantial progressby representing novel class objects using the feature representation learned upon a set of base class objects. However,an implicit contradiction between novel class classification and representation is unfortunately ignored. On the one hand, to achieve accurate novel class classification, the distributions of either two base classes must be far away fromeach other (max-margin). On the other hand, to precisely represent novel classes, the distributions of base classes should be close to each other to reduce the intra-class distance of novel classes (min-margin). In this paper, we propose a class margin equilibrium (CME) approach, with the aim to optimize both feature space partition and novel class reconstruction in a systematic way. CME first converts the few-shot detection problem to the few-shot classification problem by using a fully connected layer to decouple localization features. CME then reserves adequate margin space for novel classes by introducing simple-yet-effective class margin loss during feature learning. Finally, CME pursues margin equilibrium by disturbing the features of novel class instances in an adversarial min-max fashion. Experiments on Pascal VOC and MS-COCO datasets show that CME significantly improves upon two baseline detectors (up to 3 ∼ 5% in average), achieving state-of-the-art performance. Code is available at https://github.com/Bohao-Lee/CME .