Tiny objects, frequently appearing in practical applications, have weak appearance and features, and receive increasing interests in meany vision tasks, such as object detection and segmentation. To promote the research and development of tiny object tracking, we create a large-scale video dataset, which contains 434 sequences with a total of more than 217K frames. Each frame is carefully annotated with a high-quality bounding box. In data creation, we take 12 challenge attributes into account to cover a broad range of viewpoints and scene complexities, and annotate these attributes for facilitating the attribute-based performance analysis. To provide a strong baseline in tiny object tracking, we propose a novel Multilevel Knowledge Distillation Network (MKDNet), which pursues three-level knowledge distillations in a unified framework to effectively enhance the feature representation, discrimination and localization abilities in tracking tiny objects. Extensive experiments are performed on the proposed dataset, and the results prove the superiority and effectiveness of MKDNet compared with state-of-the-art methods. The dataset, the algorithm code, and the evaluation code are available at https://github.com/mmic-lcl/Datasets-and-benchmark-code.
In this work we present point-level region contrast, a self-supervised pre-training approach for the task of object detection. This approach is motivated by the two key factors in detection: localization and recognition. While accurate localization favors models that operate at the pixel- or point-level, correct recognition typically relies on a more holistic, region-level view of objects. Incorporating this perspective in pre-training, our approach performs contrastive learning by directly sampling individual point pairs from different regions. Compared to an aggregated representation per region, our approach is more robust to the change in input region quality, and further enables us to implicitly improve initial region assignments via online knowledge distillation during training. Both advantages are important when dealing with imperfect regions encountered in the unsupervised setting. Experiments show point-level region contrast improves on state-of-the-art pre-training methods for object detection and segmentation across multiple tasks and datasets, and we provide extensive ablation studies and visualizations to aid understanding. Code will be made available.
Knowledge Distillation has shown very promising abil-ity in transferring learned representation from the largermodel (teacher) to the smaller one (student).Despitemany efforts, prior methods ignore the important role ofretaining inter-channel correlation of features, leading tothe lack of capturing intrinsic distribution of the featurespace and sufficient diversity properties of features in theteacher network.To solve the issue, we propose thenovel Inter-Channel Correlation for Knowledge Distillation(ICKD), with which the diversity and homology of the fea-ture space of the student network can align with that ofthe teacher network. The correlation between these twochannels is interpreted as diversity if they are irrelevantto each other, otherwise homology. Then the student isrequired to mimic the correlation within its own embed-ding space. In addition, we introduce the grid-level inter-channel correlation, making it capable of dense predictiontasks. Extensive experiments on two vision tasks, includ-ing ImageNet classification and Pascal VOC segmentation,demonstrate the superiority of our ICKD, which consis-tently outperforms many existing methods, advancing thestate-of-the-art in the fields of Knowledge Distillation. Toour knowledge, we are the first method based on knowl-edge distillation boosts ResNet18 beyond 72% Top-1 ac-curacy on ImageNet classification. Code is available at:https://github.com/ADLab-AutoDrive/ICKD.
In several supervised learning scenarios, auxiliary losses are used in order to introduce additional information or constraints into the supervised learning objective. For instance, knowledge distillation aims to mimic outputs of a powerful teacher model; similarly, in rule-based approaches, weak labeling information is provided by labeling functions which may be noisy rule-based approximations to true labels. We tackle the problem of learning to combine these losses in a principled manner. Our proposal, AMAL, uses a bi-level optimization criterion on validation data to learn optimal mixing weights, at an instance level, over the training data. We describe a meta-learning approach towards solving this bi-level objective and show how it can be applied to different scenarios in supervised learning. Experiments in a number of knowledge distillation and rule-denoising domains show that AMAL provides noticeable gains over competitive baselines in those domains. We empirically analyze our method and share insights into the mechanisms through which it provides performance gains.
Deep learning often requires a large amount of data. In real-world applications, e.g., healthcare applications, the data collected by a single organization (e.g., hospital) is often limited, and the majority of massive and diverse data is often segregated across multiple organizations. As such, it motivates the researchers to conduct distributed deep learning, where the data user would like to build DL models using the data segregated across multiple different data owners. However, this could lead to severe privacy concerns due to the sensitive nature of the data, thus the data owners would be hesitant and reluctant to participate. We propose LDP-DL, a privacy-preserving distributed deep learning framework via local differential privacy and knowledge distillation, where each data owner learns a teacher model using its own (local) private dataset, and the data user learns a student model to mimic the output of the ensemble of the teacher models. In the experimental evaluation, a comprehensive comparison has been made among our proposed approach (i.e., LDP-DL), DP-SGD, PATE and DP-FL, using three popular deep learning benchmark datasets (i.e., CIFAR10, MNIST and FashionMNIST). The experimental results show that LDP-DL consistently outperforms the other competitors in terms of privacy budget and model accuracy.
LoFTR arXiv:2104.00680 is an efficient deep learning method for finding appropriate local feature matches on image pairs. This paper reports on the optimization of this method to work on devices with low computational performance and limited memory. The original LoFTR approach is based on a ResNet arXiv:1512.03385 head and two modules based on Linear Transformer arXiv:2006.04768 architecture. In the presented work, only the coarse-matching block was left, the number of parameters was significantly reduced, and the network was trained using a knowledge distillation technique. The comparison showed that this approach allows to obtain an appropriate feature detection accuracy for the student model compared to the teacher model in the coarse matching block, despite the significant reduction of model size. Also, the paper shows additional steps required to make model compatible with NVIDIA TensorRT runtime, and shows an approach to optimize training method for low-end GPUs.
Social media plays a significant role in sharing essential information, which helps humanitarian organizations in rescue operations during and after disaster incidents. However, developing an efficient method that can provide rapid analysis of social media images in the early hours of disasters is still largely an open problem, mainly due to the lack of suitable datasets and the sheer complexity of this task. In addition, supervised methods can not generalize well to novel disaster incidents. In this paper, inspired by the success of Knowledge Distillation (KD) methods, we propose an unsupervised deep neural network to detect and localize damages in social media images. Our proposed KD architecture is a feature-based distillation approach that comprises a pre-trained teacher and a smaller student network, with both networks having similar GAN architecture containing a generator and a discriminator. The student network is trained to emulate the behavior of the teacher on training input samples, which, in turn, contain images that do not include any damaged regions. Therefore, the student network only learns the distribution of no damage data and would have different behavior from the teacher network-facing damages. To detect damage, we utilize the difference between features generated by two networks using a defined score function that demonstrates the probability of damages occurring. Our experimental results on the benchmark dataset confirm that our approach outperforms state-of-the-art methods in detecting and localizing the damaged areas, especially for novel disaster types.
Despite their success, deep networks have been shown to be highly susceptible to perturbations, often causing significant drops in accuracy. In this paper, we investigate model robustness on perturbed inputs by studying the performance of internal sub-networks (subnets). Interestingly, we observe that most subnets show particularly poor robustness against perturbations. More importantly, these weak subnets are correlated with the overall lack of robustness. Tackling this phenomenon, we propose a new training procedure that identifies and enhances weak subnets (EWS) to improve robustness. Specifically, we develop a search algorithm to find particularly weak subnets and explicitly strengthen them via knowledge distillation from the full network. We show that EWS greatly improves both robustness against corrupted images as well as accuracy on clean data. Being complementary to popular data augmentation methods, EWS consistently improves robustness when combined with these approaches. To highlight the flexibility of our approach, we combine EWS also with popular adversarial training methods resulting in improved adversarial robustness.
Global and local relational reasoning enable scene understanding models to perform human-like scene analysis and understanding. Scene understanding enables better semantic segmentation and object-to-object interaction detection. In the medical domain, a robust surgical scene understanding model allows the automation of surgical skill evaluation, real-time monitoring of surgeon's performance and post-surgical analysis. This paper introduces a globally-reasoned multi-task surgical scene understanding model capable of performing instrument segmentation and tool-tissue interaction detection. Here, we incorporate global relational reasoning in the latent interaction space and introduce multi-scale local (neighborhood) reasoning in the coordinate space to improve segmentation. Utilizing the multi-task model setup, the performance of the visual-semantic graph attention network in interaction detection is further enhanced through global reasoning. The global interaction space features from the segmentation module are introduced into the graph network, allowing it to detect interactions based on both node-to-node and global interaction reasoning. Our model reduces the computation cost compared to running two independent single-task models by sharing common modules, which is indispensable for practical applications. Using a sequential optimization technique, the proposed multi-task model outperforms other state-of-the-art single-task models on the MICCAI endoscopic vision challenge 2018 dataset. Additionally, we also observe the performance of the multi-task model when trained using the knowledge distillation technique. The official code implementation is made available in GitHub.
Knowledge Distillation is a technique which aims to utilize dark knowledge to compress and transfer information from a vast, well-trained neural network (teacher model) to a smaller, less capable neural network (student model) with improved inference efficiency. This approach of distilling knowledge has gained popularity as a result of the prohibitively complicated nature of such cumbersome models for deployment on edge computing devices. Generally, the teacher models used to teach smaller student models are cumbersome in nature and expensive to train. To eliminate the necessity for a cumbersome teacher model completely, we propose a simple yet effective knowledge distillation framework that we termed Dynamic Rectification Knowledge Distillation (DR-KD). Our method transforms the student into its own teacher, and if the self-teacher makes wrong predictions while distilling information, the error is rectified prior to the knowledge being distilled. Specifically, the teacher targets are dynamically tweaked by the agency of ground-truth while distilling the knowledge gained from traditional training. Our proposed DR-KD performs remarkably well in the absence of a sophisticated cumbersome teacher model and achieves comparable performance to existing state-of-the-art teacher-free knowledge distillation frameworks when implemented by a low-cost dynamic mannered teacher. Our approach is all-encompassing and can be utilized for any deep neural network training that requires categorization or object recognition. DR-KD enhances the test accuracy on Tiny ImageNet by 2.65% over prominent baseline models, which is significantly better than any other knowledge distillation approach while requiring no additional training costs.
Knowledge distillation (KD) achieves promising results on the challenging problem of unsupervised anomaly detection (AD).The representation discrepancy of anomalies in the teacher-student (T-S) model provides essential evidence for AD. However, using similar or identical architectures to build the teacher and student models in previous studies hinders the diversity of anomalous representations. To tackle this problem, we propose a novel T-S model consisting of a teacher encoder and a student decoder and introduce a simple yet effective "reverse distillation" paradigm accordingly. Instead of receiving raw images directly, the student network takes teacher model's one-class embedding as input and targets to restore the teacher's multiscale representations. Inherently, knowledge distillation in this study starts from abstract, high-level presentations to low-level features. In addition, we introduce a trainable one-class bottleneck embedding (OCBE) module in our T-S model. The obtained compact embedding effectively preserves essential information on normal patterns, but abandons anomaly perturbations. Extensive experimentation on AD and one-class novelty detection benchmarks shows that our method surpasses SOTA performance, demonstrating our proposed approach's effectiveness and generalizability.
Multitask learning (MTL) has recently gained a lot of popularity as a learning paradigm that can lead to improved per-task performance while also using fewer per-task model parameters compared to single task learning. One of the biggest challenges regarding MTL networks involves how to share features across tasks. To address this challenge, we propose the Attentive Task Interaction Network (ATI-Net). ATI-Net employs knowledge distillation of the latent features for each task, then combines the feature maps to provide improved contextualized information to the decoder. This novel approach to introducing knowledge distillation into an attention based multitask network outperforms state of the art MTL baselines such as the standalone MTAN and PAD-Net, with roughly the same number of model parameters.
Representation knowledge distillation aims at transferring rich information from one model to another. Common approaches for representation distillation mainly focus on the direct minimization of distance metrics between the models' embedding vectors. Such direct methods may be limited in transferring high-order dependencies embedded in the representation vectors, or in handling the capacity gap between the teacher and student models. Moreover, in standard knowledge distillation, the teacher is trained without awareness of the student's characteristics and capacity. In this paper, we explore two mechanisms for enhancing representation distillation using classifier sharing between the teacher and student. We first investigate a simple scheme where the teacher's classifier is connected to the student backbone, acting as an additional classification head. Then, we propose a student-aware mechanism that asks to tailor the teacher model to a student with limited capacity by training the teacher with a temporary student's head. We analyze and compare these two mechanisms and show their effectiveness on various datasets and tasks, including image classification, fine-grained classification, and face verification. In particular, we achieve state-of-the-art results for face verification on the IJB-C dataset for a MobileFaceNet model: TAR@(FAR=1e-5)=93.7\%. Code is available at https://github.com/Alibaba-MIIL/HeadSharingKD.
Deep neural networks are typically trained in a single shot for a specific task and data distribution, but in real world settings both the task and the domain of application can change. The problem becomes even more challenging in dense predictive tasks, such as semantic segmentation, and furthermore most approaches tackle the two problems separately. In this paper we introduce the novel task of coarse-to-fine learning of semantic segmentation architectures in presence of domain shift. We consider subsequent learning stages progressively refining the task at the semantic level; i.e., the finer set of semantic labels at each learning step is hierarchically derived from the coarser set of the previous step. We propose a new approach (CCDA) to tackle this scenario. First, we employ the maximum squares loss to align source and target domains and, at the same time, to balance the gradients between well-classified and harder samples. Second, we introduce a novel coarse-to-fine knowledge distillation constraint to transfer network capabilities acquired on a coarser set of labels to a set of finer labels. Finally, we design a coarse-to-fine weight initialization rule to spread the importance from each coarse class to the respective finer classes. To evaluate our approach, we design two benchmarks where source knowledge is extracted from the GTA5 dataset and it is transferred to either the Cityscapes or the IDD datasets, and we show how it outperforms the main competitors.
Feature regression is a simple way to distill large neural network models to smaller ones. We show that with simple changes to the network architecture, regression can outperform more complex state-of-the-art approaches for knowledge distillation from self-supervised models. Surprisingly, the addition of a multi-layer perceptron head to the CNN backbone is beneficial even if used only during distillation and discarded in the downstream task. Deeper non-linear projections can thus be used to accurately mimic the teacher without changing inference architecture and time. Moreover, we utilize independent projection heads to simultaneously distill multiple teacher networks. We also find that using the same weakly augmented image as input for both teacher and student networks aids distillation. Experiments on ImageNet dataset demonstrate the efficacy of the proposed changes in various self-supervised distillation settings.
Visible-infrared person re-identification (VI-ReID) has been challenging due to the existence of large discrepancies between visible and infrared modalities. Most pioneering approaches reduce intra-class variations and inter-modality discrepancies by learning modality-shared and ID-related features. However, an explicit modality-shared cue, i.e., body keypoints, has not been fully exploited in VI-ReID. Additionally, existing feature learning paradigms imposed constraints on either global features or partitioned feature stripes, which neglect the prediction consistency of global and part features. To address the above problems, we exploit Pose Estimation as an auxiliary learning task to assist the VI-ReID task in an end-to-end framework. By jointly training these two tasks in a mutually beneficial manner, our model learns higher quality modality-shared and ID-related features. On top of it, the learnings of global features and local features are seamlessly synchronized by Hierarchical Feature Constraint (HFC), where the former supervises the latter using the knowledge distillation strategy. Experimental results on two benchmark VI-ReID datasets show that the proposed method consistently improves state-of-the-art methods by significant margins. Specifically, our method achieves nearly 20$\%$ mAP improvements against the state-of-the-art method on the RegDB dataset. Our intriguing findings highlight the usage of auxiliary task learning in VI-ReID.
Advanced face swapping methods have achieved appealing results. However, most of these methods have many parameters and computations, which makes it challenging to apply them in real-time applications or deploy them on edge devices like mobile phones. In this work, we propose a lightweight Identity-aware Dynamic Network (IDN) for subject-agnostic face swapping by dynamically adjusting the model parameters according to the identity information. In particular, we design an efficient Identity Injection Module (IIM) by introducing two dynamic neural network techniques, including the weights prediction and weights modulation. Once the IDN is updated, it can be applied to swap faces given any target image or video. The presented IDN contains only 0.50M parameters and needs 0.33G FLOPs per frame, making it capable for real-time video face swapping on mobile phones. In addition, we introduce a knowledge distillation-based method for stable training, and a loss reweighting module is employed to obtain better synthesized results. Finally, our method achieves comparable results with the teacher models and other state-of-the-art methods.
Data-Free Knowledge Distillation (KD) allows knowledge transfer from a trained neural network (teacher) to a more compact one (student) in the absence of original training data. Existing works use a validation set to monitor the accuracy of the student over real data and report the highest performance throughout the entire process. However, validation data may not be available at distillation time either, making it infeasible to record the student snapshot that achieved the peak accuracy. Therefore, a practical data-free KD method should be robust and ideally provide monotonically increasing student accuracy during distillation. This is challenging because the student experiences knowledge degradation due to the distribution shift of the synthetic data. A straightforward approach to overcome this issue is to store and rehearse the generated samples periodically, which increases the memory footprint and creates privacy concerns. We propose to model the distribution of the previously observed synthetic samples with a generative network. In particular, we design a Variational Autoencoder (VAE) with a training objective that is customized to learn the synthetic data representations optimally. The student is rehearsed by the generative pseudo replay technique, with samples produced by the VAE. Hence knowledge degradation can be prevented without storing any samples. Experiments on image classification benchmarks show that our method optimizes the expected value of the distilled model accuracy while eliminating the large memory overhead incurred by the sample-storing methods.
Deep reinforcement learning has shown promising results on an abundance of robotic tasks in simulation, including visual navigation and manipulation. Prior work generally aims to build embodied agents that solve their assigned tasks as quickly as possible, while largely ignoring the problems caused by collision with objects during interaction. This lack of prioritization is understandable: there is no inherent cost in breaking virtual objects. As a result, "well-trained" agents frequently collide with objects before achieving their primary goals, a behavior that would be catastrophic in the real world. In this paper, we study the problem of training agents to complete the task of visual mobile manipulation in the ManipulaTHOR environment while avoiding unnecessary collision (disturbance) with objects. We formulate disturbance avoidance as a penalty term in the reward function, but find that directly training with such penalized rewards often results in agents being unable to escape poor local optima. Instead, we propose a two-stage training curriculum where an agent is first allowed to freely explore and build basic competencies without penalization, after which a disturbance penalty is introduced to refine the agent's behavior. Results on testing scenes show that our curriculum not only avoids these poor local optima, but also leads to 10% absolute gains in success rate without disturbance, compared to our state-of-the-art baselines. Moreover, our curriculum is significantly more performant than a safe RL algorithm that casts collision avoidance as a constraint. Finally, we propose a novel disturbance-prediction auxiliary task that accelerates learning.
The staple of human intelligence is the capability of acquiring knowledge in a continuous fashion. In stark contrast, Deep Networks forget catastrophically and, for this reason, the sub-field of Class-Incremental Continual Learning fosters methods that learn a sequence of tasks incrementally, blending sequentially-gained knowledge into a comprehensive prediction. This work aims at assessing and overcoming the pitfalls of our previous proposal Dark Experience Replay (DER), a simple and effective approach that combines rehearsal and Knowledge Distillation. Inspired by the way our minds constantly rewrite past recollections and set expectations for the future, we endow our model with the abilities to i) revise its replay memory to welcome novel information regarding past data ii) pave the way for learning yet unseen classes. We show that the application of these strategies leads to remarkable improvements; indeed, the resulting method - termed eXtended-DER (X-DER) - outperforms the state of the art on both standard benchmarks (such as CIFAR-100 and miniImagenet) and a novel one here introduced. To gain a better understanding, we further provide extensive ablation studies that corroborate and extend the findings of our previous research (e.g. the value of Knowledge Distillation and flatter minima in continual learning setups).