Knowledge distillation has been applied to various tasks successfully. The current distillation algorithm usually improves students’ performance by imitating the output of the teacher. This paper shows that teachers can also improve students’ representation power by guiding students’ feature recovery. From this point of view, we propose Masked Generative Distillation (MGD), which is simple: we mask random pixels of the student’s feature and force it to generate the teacher’s full feature through a simple block. MGD is a truly general feature-based distillation method, which can be utilized on various tasks, including image classification, object detection, semantic segmentation and instance segmentation. We experiment on different models with extensive datasets and the results show that all the students achieve excellent improvements. Notably, we boost ResNet-18 from 69.90% to 71.69% ImageNet top-1 accuracy, RetinaNet with ResNet-50 backbone from 37.4 to 41.0 Boundingbox mAP, SOLO based on ResNet-50 from 33.1 to 36.2 Mask mAP and DeepLabV3 based on ResNet-18 from 73.20 to 76.02 mIoU. Our codes are available at https://github.com/yzd-v/MGD.
论文检索
输入标题、作者或关键词，从 1,645 篇学术成果中精准定位
The Spiking Neural Network (SNN) has attracted more and more attention recently. It adopts binary spike signals to transmit information. Benefitting from the information passing paradigm of SNNs, the multiplications of activations and weights can be replaced by additions, which are more energy-efficient. However, its “Hard Reset"" mechanism for the firing activity would ignore the difference among membrane potentials when the membrane potential is above the firing threshold, causing information loss. Meanwhile, quantifying the membrane potential to 0/1 spikes at the firing instants will inevitably introduce the quantization error thus bringing about information loss too. To address these problems, we propose a “Soft Reset"" mechanism for the supervised training-based SNNs, which will drive the membrane potential to a dynamic reset potential according to its magnitude, and Membrane Potential Rectifier (MPR) to reduce the quantization error via redistributing the membrane potential to a range close to the spikes. Experimental results show that the SNNs with the proposed “Soft Reset"" mechanism and MPR outperform their vanilla counterparts on both static and dynamic datasets.
Knowledge Distillation (KD) transfers the knowledge from a high-capacity teacher network to strengthen a smaller student. Existing methods focus on excavating the knowledge hints and transferring the whole knowledge to the student. However, the knowledge redundancy arises since the knowledge shows different values to the student at different learning stages. In this paper, we propose Knowledge Condensation Distillation (KCD). Specifically, the knowledge value on each sample is dynamically estimated, based on which an Expectation-Maximization (EM) framework is forged to iteratively condense a compact knowledge set from the teacher to guide the student learning. Our approach is easy to build on top of the off-the-shelf KD methods, with no extra training parameters and negligible computation overhead. Thus, it presents one new perspective for KD, in which the student that actively identifies teacher’s knowledge in line with its aptitude can learn to learn more effectively and efficiently. Experiments on standard benchmarks manifest that the proposed KCD can well boost the performance of student model with even higher distillation efficiency. Code is available at https://github.com/dzy3/KCD.
Current object detection systems and benchmarks typically handle a limited number of categories, up to about a thousand categories. This paper scales the number of categories for object detection systems and benchmarks up to 21,000, by leveraging existing object detection and image classification data. Unlike previous efforts that usually transfer knowledge from base detectors to image classification data, we propose to rely more on a reverse information flow from a base image classifier to object detection data. In this framework, the large-vocabulary classification capability is first learnt thoroughly using only the image classification data. In this step, the image classification problem is reformulated as a special configuration of object detection that treats the entire image as a special RoI. Then, a simple multi-task learning approach is used to join the image classification and object detection data, with the backbone and the RoI classification branch shared between two tasks. This two-stage approach, though very simple without a sophisticated process such as multi-instance learning (MIL) to generate pseudo labels for object proposals on the image classification data, performs rather strongly that it surpasses previous large-vocabulary object detection systems on a standard evaluation protocol of tailored LVIS. Considering that the tailored LVIS evaluation only accounts for a few hundred novel object categories, we present a new evaluation benchmark that assesses the detection of all 21,841 object classes in the ImageNet-21K dataset. The baseline approach and evaluation benchmark will be publicly available at https://github.com/SwinTransformer/Simple-21K-Detection. We hope these would ease future research on large-vocabulary object detection.
Combining simple architectures with large-scale pre-training has led to massive improvements in image classification. For object detection, pre-training and scaling approaches are less well established, especially in the long-tailed and open-vocabulary setting, where training data is relatively scarce. In this paper, we propose a strong recipe for transferring image-text models to open-vocabulary object detection. We use a standard Vision Transformer architecture with minimal modifications, contrastive image-text pre-training, and end-to-end detection fine-tuning. Our analysis of the scaling properties of this setup shows that increasing image-level pre-training and model size yield consistent improvements on the downstream detection task. We provide the adaptation strategies and regularizations needed to attain very strong performance on zero-shot text-conditioned and one-shot image-conditioned object detection. Code and models will be made available on GitHub.
This work presents a simple vision transformer design as a strong baseline for object localization and instance segmentation tasks. Transformers recently demonstrate competitive performance in image classification tasks. To adopt ViT to object detection and dense prediction tasks, many works inherit the multistage design from convolutional networks and highly customized ViT architectures. Behind this design, the goal is to pursue a better trade-off between computational cost and effective aggregation of multiscale global contexts. However, existing works adopt multistage architectural design as a black-box solution without a clear understanding of its true benefits. In this paper, we comprehensively study three architecture design choices on ViT -- spatial reduction, doubled channels, and multiscale features -- and demonstrate that a vanilla ViT architecture can fulfill this goal without handcrafting multiscale features, maintaining the original ViT design philosophy. We further complete a scaling rule to optimize our model’s trade-off on accuracy and computation cost / model size. By leveraging a constant feature resolution and hidden size throughout the encoder blocks, we propose a simple and compact ViT architecture called Universal Vision Transformer (UViT) that achieves strong performance on COCO object detection and instance segmentation benchmark. Our code will be released upon acceptance.
Although deep-learning based methods for monocular pedestrian detection have made a great progress, they are still vulnerable to heavy occlusions. Using multi-view information fusion is a potential solution but has limited applications, due to the lack of annotated training samples in existing multi-view datasets, which increases the risk of overfitting. To address this problem, a data augmentation method is proposed to randomly generate 3D cylinder occlusions, on the ground plane, which are of the average size of pedestrians and projected to multiple views, to relieve the impact of overfitting in the training. Moreover, the feature map of each view is projected to multiple parallel planes at different heights, by using homographies, which allows the CNNs to fully utilize the features across the height of each pedestrian to infer the locations of pedestrians on the ground plane. The proposed 3DROM method has a greatly improved performance in comparison with the state-of-the-art deep-learning based methods for multi-view pedestrian detection.
3D bounding boxes are a widespread intermediate representation in many computer vision applications. However, predicting them is a challenging task, largely due to partial observability, which motivates the need for a strong sense of uncertainty. While many recent methods have explored better architectures for consuming sparse and unstructured point cloud data, we hypothesize that there is room for improvement in the modeling of the output distribution and explore how this can be achieved using an autoregressive prediction head. Additionally, we release a simulated dataset, COB-3D, which highlights new types of ambiguity that arise in real-world robotics applications, where 3D bounding box prediction has largely been underexplored. We propose methods for leveraging our autoregressive model to make high confidence predictions and meaningful uncertainty measures, achieving strong results on SUN-RGBD, Scannet, KITTI, and our new dataset. Code and dataset are available at https://bbox.yuxuanliu.com.
One fundamental challenge in building an instance segmentation model for a large number of classes in complex scenes is the lack of training examples, especially for rare objects. In this paper, we explore the possibility to increase the training examples without laborious data collection and annotation. We find that an abundance of instance segments can potentially be obtained freely from object-centric images, according to two insights: (i) an object-centric image usually contains one salient object in a simple background; (ii) objects from the same class often share similar appearances or similar contrasts to the background. Motivated by these insights, we propose a simple and scalable framework FreeSeg for extracting and leveraging these ""free"" object foreground segments to facilitate model training in long-tailed instance segmentation. Concretely, we investigate the similarity among object-centric images of the same class to propose candidate segments of foreground instances, followed by a novel ranking of segment quality. The resulting high-quality object segments can then be used to augment the existing long-tailed datasets, e.g., by copying and pasting the segments onto the original training images. Extensive experiments show that FreeSeg yields substantial improvements on top of strong baselines and achieves state-of-the-art accuracy for segmenting rare object categories. Our code is publicly available at https://github.com/czhang0528/FreeSeg.
The superior performance of object detectors is often established under the condition that the test samples are in the same distribution as the training data. However, in most practical applications, out-of-distribution (OOD) instances are inevitable and usually lead to detection uncertainty. In this work, the Feature structured OOD-IDentification (FOOD-ID) model is proposed to reduce the uncertainty of detection results by identifying the OOD instances. Instead of outputting each detection result directly, FOOD-ID uses a likelihood-based measuring mechanism to identify whether the feature satisfies the corresponding class distribution and outputs the OOD results separately. Specifically, the clustering-oriented feature structuration is firstly developed using class-specified prototypes and Attractive-Repulsive loss for more discriminative feature representation and more compact distribution. With the structured features space, the density distribution of all training categories is estimated based on a class-conditional normalizing flow, which is then used for the OOD identification in the test stage. The proposed FOOD-ID can be easily applied to various object detectors including anchor-based frameworks and anchor-free frameworks. Extensive experiments on the PASCAL VOC dataset and an industrial defect dataset demonstrate that FOOD-ID achieves satisfactory OOD identification performance, with which the certainty of detection results is improved significantly.
We address the task of person search, that is, localizing and re-identifying query persons from a set of raw scene images. Recent approaches are typically built upon OIMNet, a pioneer work on person search, that learns joint person representations for performing both detection and person re-identification (reID) tasks. To obtain the representations, they extract features from pedestrian proposals, and then project them on a unit hypersphere with L2 normalization. These methods also incorporate all positive proposals, that sufficiently overlap with the ground truth, equally to learn person representations for reID. We have found that 1) the L2 normalization without considering feature distributions degenerates the discriminative power of person representations, and 2) positive proposals often also depict background clutter and person overlaps, which could encode noisy features to person representations. In this paper, we introduce OIMNet++ that addresses the aforementioned limitations. To this end, we introduce a novel normalization layer, dubbed ProtoNorm, that calibrates features from pedestrian proposals, while considering a long-tail distribution of person IDs, enabling L2 normalized person representations to be discriminative. We also propose a localization-aware feature learning scheme that encourages better-aligned proposals to contribute more in learning discriminative representations. Experimental results and analysis on standard person search benchmarks demonstrate the effectiveness of OIMNet++.
Weakly supervised object localization (WSOL) aims to learn representations that encode object location using only image-level category labels. However, many objects can be labeled at different levels of granularity. Is it an animal, a bird, or a great horned owl? Which image-level labels should we use? In this paper we study the role of label granularity in WSOL. To facilitate this investigation we introduce iNatLoc500, a new large-scale fine-grained benchmark dataset for WSOL. Surprisingly, we find that choosing the right training label granularity provides a much larger performance boost than choosing the best WSOL algorithm. We also show that changing the label granularity can significantly improve data efficiency.
We propose a method for estimating the 6DoF pose of a rigid object with an available 3D model from a single RGB image. Unlike classical correspondence-based methods which predict 3D object coordinates at pixels of the input image, the proposed method predicts 3D object coordinates at 3D query points sampled in the camera frustum. The move from pixels to 3D points, which is inspired by recent PIFu-style methods for 3D reconstruction, enables reasoning about the whole object, including its (self-)occluded parts. For a 3D query point associated with a pixel-aligned image feature, we train a fully-connected neural network to predict: (i) the corresponding 3D object coordinates, and (ii) the signed distance to the object surface, with the first defined only for query points in the surface vicinity. We call the mapping realized by this network as Neural Correspondence Field. The object pose is then robustly estimated from the predicted 3D-3D correspondences by the Kabsch-RANSAC algorithm. The proposed method achieves state-of-the-art results on three BOP datasets and is shown superior especially in challenging cases with occlusion. The project website is at: linhuang17.github.io/NCF.
Weakly-Supervised Temporal Action Detection for Fine-Grained Videos with Hierarchical Atomic Actions
PDF ↗Action understanding has evolved into the era of fine granularity, as most human behaviors in real life have only minor differences. To detect these fine-grained actions accurately in a label-efficient way, we tackle the problem of weakly-supervised fine-grained temporal action detection in videos for the first time. Without the careful design to capture subtle differences between fine-grained actions, previous weakly-supervised models for general action detection cannot perform well in fine-grained settings. We propose to model actions as the combinations of reusable atomic actions which are automatically discovered from data through self-supervised clustering, in order to capture the commonality and individuality of fine-grained actions. The learnt atomic actions, represented by visual concepts, are further mapped to fine and coarse action labels leveraging the semantic label hierarchy. Our approach constructs a visual representation hierarchy of four levels: clip level, atomic action level, fine action class level and coarse action class level, with supervision at each level. Extensive experiments on two large-scale fine-grained video datasets, FineAction and FineGym, show the benefit of our proposed weakly-supervised model for fine-grained action detection, and it achieves state-of-the-art results.
Accurate uncertainty estimates are essential for deploying deep object detectors in safety-critical systems. The development and evaluation of probabilistic object detectors have been hindered by shortcomings in existing performance measures, which tend to involve arbitrary thresholds or limit the detector’s choice of distributions. In this work, we propose to view object detection as a set prediction task where detectors predict the distribution over the set of objects. Using the negative log-likelihood for random finite sets, we present a proper scoring rule for evaluating and training probabilistic object detectors. The proposed method can be applied to existing probabilistic detectors, is free from thresholds, and enables fair comparison between architectures. Three different types of detectors are evaluated on the COCO dataset. Our results indicate that the training of existing detectors is optimized toward non-probabilistic metrics. We hope to encourage the development of new object detectors that can accurately estimate their own uncertainty.
LiDAR and Camera sensors have complementary properties: LiDAR senses accurate positioning, while camera provides rich texture and color information. Fusing these two modalities can intuitively improve the performance of 3D detection. Most multi-modal fusion methods use networks to extract features of LiDAR and camera modality respectively, then simply add or concancate them together. We argue that these two kinds of signals are completely different, so it is not proper to combine these two heterogeneous features directly. In this paper, we propose EMMF-Det to do multi-modal fusion leveraging range and camera images. EMMF-Det uses self-attention mechanism to do feature re-weighting on these two modalities interactively, which can enchance the features with color, texture and localiztion information provided by LiDAR and camera signals. On the Waymo Open Dataset, EMMF-Det acheives the state-of-the-art performance. Besides this, evaluation on self-built dataset further proves the effectiveness of our method.
What constitutes an object? This has been a long-standing question in computer vision. Towards this goal, numerous learning-free and learning-based approaches have been developed to score objectness. However, they generally do not scale well across new domains and for unseen objects. In this paper, we advocate that existing methods lack a top-down supervision signal governed by human-understandable semantics. For the first time in literature, we demonstrate that Multi-modal Vision Transformers (MViT) trained with aligned image-text pairs can effectively bridge this gap. Our extensive experiments across various domains and novel objects show the state-of-the-art performance of MViTs to localize generic objects in images. Based on the observation that existing MViTs do not include multi-scale feature processing and usually require longer training schedules, we develop an efficient and flexible MViT architecture using multi-scale and deformable self-attention. We show the significance of MViT proposals in a diverse range of applications including open-world object detection, salient and camouflage object detection, supervised and self-supervised detection tasks. Further, MViTs can adaptively generate proposals given a specific language query and thus offer enhanced interactability. Code: https://git.io/J1HPY.
Video Anomaly Detection (VAD) is an important topic in computer vision. Motivated by the recent advances in self-supervised learning, this paper addresses VAD by solving an intuitive yet challenging pretext task, i.e., spatio-temporal jigsaw puzzles, which is cast as a multi-label fine-grained classification problem. Our method exhibits several advantages over existing works: 1) the spatio-temporal jigsaw puzzles are decoupled in terms of spatial and temporal dimensions, responsible for capturing highly discriminative appearance and motion features, respectively; 2) full permutations are used to provide abundant jigsaw puzzles covering various difficulty levels, allowing the network to distinguish subtle spatio-temporal differences between normal and abnormal events; and 3) the pretext task is tackled in an end-to-end manner without relying on any pre-trained models. Our method outperforms state-of-the-art counterparts on three public benchmarks. Especially on ShanghaiTech Campus, the result is superior to reconstruction and prediction-based methods by a large margin.
Recently, promising applications in robotics and augmented reality have attracted considerable attention to 3D object detection from point clouds. In this paper, we present FCAF3D -- a first-in-class fully convolutional anchor-free indoor 3D object detection method. It is a simple yet effective method that uses a voxel representation of a point cloud and processes voxels with sparse convolutions. FCAF3D can handle large-scale scenes with minimal runtime through a single fully convolutional feed-forward pass. Existing 3D object detection methods make prior assumptions on the geometry of objects, and we argue that it limits their generalization ability. To eliminate prior assumptions, we propose a novel parametrization of oriented bounding boxes that allows obtaining better results in a purely data-driven way. The proposed method achieves state-of-the-art 3D object detection results in terms of mAP@0.5 on ScanNet V2 (+4.5), SUN RGB-D (+3.5), and S3DIS (+20.5) datasets.The code and models are available at https://github.com/samsunglabs/fcaf3d
Knowledge distillation (KD) is a well-known model compression strategy to improve models’ performance with fewer parameters. However, recent KD approaches for object detection have faced two limitations. First, they distill nearby foreground regions, ignoring potentially useful background information. Second, they only consider global contexts, thereby the student model can hardly learn local details from the teacher model. To overcome such challenging issues, we propose a novel knowledge distillation method, GLAMD, distilling both global and local knowledge from the teacher. We divide the feature maps into several patches and apply an attention mechanism for both the entire feature area and each patch to extract the global context as well as local details simultaneously. Our method outperforms the state-of-the-art methods with 40.8 AP on COCO2017 dataset, which is 3.4 AP higher than the student model (ResNet50 based Faster R-CNN) and 0.7 AP higher than the previous global attention-based distillation method.