Recently, Masked Image Modeling (MIM) achieves great success in self-supervised visual recognition. However, as a reconstruction-based framework, it is still an open question to understand how MIM works, since MIM appears very different from previous well-studied siamese approaches such as contrastive learning. In this paper, we propose a new viewpoint: MIM implicitly learns occlusion-invariant features, which is analogous to other siamese methods while the latter learns other invariance. By relaxing MIM formulation into an equivalent siamese form, MIM methods can be interpreted in a unified framework with conventional methods, among which only a) data transformations, i.e. what invariance to learn, and b) similarity measurements are different. Furthermore, taking MAE (He et al.) as a representative example of MIM, we empirically find the success of MIM models relates a little to the choice of similarity functions, but the learned occlusion invariant feature introduced by masked image -- it turns out to be a favored initialization for vision transformers, even though the learned feature could be less semantic. We hope our findings could inspire researchers to develop more powerful self-supervised methods in computer vision community.
We focus on better understanding the critical factors of augmentation-invariant representation learning. We revisit MoCo v2 and BYOL and try to prove the authenticity of the following assumption: different frameworks bring about representations of different characteristics even with the same pretext task. We establish the first benchmark for fair comparisons between MoCo v2 and BYOL, and observe: (i) sophisticated model configurations enable better adaptation to pre-training dataset; (ii) mismatched optimization strategies of pre-training and fine-tuning hinder model from achieving competitive transfer performances. Given the fair benchmark, we make further investigation and find asymmetry of network structure endows contrastive frameworks to work well under the linear evaluation protocol, while may hurt the transfer performances on long-tailed classification tasks. Moreover, negative samples do not make models more sensible to the choice of data augmentations, nor does the asymmetric network structure. We believe our findings provide useful information for future work.
Deep learning has substantially boosted the performance of Monocular Depth Estimation (MDE), a critical component in fully vision-based autonomous driving (AD) systems (e.g., Tesla and Toyota). In this work, we develop an attack against learning-based MDE. In particular, we use an optimization-based method to systematically generate stealthy physical-object-oriented adversarial patches to attack depth estimation. We balance the stealth and effectiveness of our attack with object-oriented adversarial design, sensitive region localization, and natural style camouflage. Using real-world driving scenarios, we evaluate our attack on concurrent MDE models and a representative downstream task for AD (i.e., 3D object detection). Experimental results show that our method can generate stealthy, effective, and robust adversarial patches for different target objects and models and achieves more than 6 meters mean depth estimation error and 93% attack success rate (ASR) in object detection with a patch of 1/9 of the vehicle's rear area. Field tests on three different driving routes with a real vehicle indicate that we cause over 6 meters mean depth estimation error and reduce the object detection rate from 90.70% to 5.16% in continuous video frames.
Videos are prone to tampering attacks that alter the meaning and deceive the audience. Previous video forgery detection schemes find tiny clues to locate the tampered areas. However, attackers can successfully evade supervision by destroying such clues using video compression or blurring. This paper proposes a video watermarking network for tampering localization. We jointly train a 3D-UNet-based watermark embedding network and a decoder that predicts the tampering mask. The perturbation made by watermark embedding is close to imperceptible. Considering that there is no off-the-shelf differentiable video codec simulator, we propose to mimic video compression by ensembling simulation results of other typical attacks, e.g., JPEG compression and blurring, as an approximation. Experimental results demonstrate that our method generates watermarked videos with good imperceptibility and robustly and accurately locates tampered areas within the attacked version.
Recent advances in 2D CNNs and vision transformers (ViTs) reveal that large kernels are essential for enough receptive fields and high performance. Inspired by this literature, we examine the feasibility and challenges of 3D large-kernel designs. We demonstrate that applying large convolutional kernels in 3D CNNs has more difficulties in both performance and efficiency. Existing techniques that work well in 2D CNNs are ineffective in 3D networks, including the popular depth-wise convolutions. To overcome these obstacles, we present the spatial-wise group convolution and its large-kernel module (SW-LK block). It avoids the optimization and efficiency issues of naive 3D large kernels. Our large-kernel 3D CNN network, i.e., LargeKernel3D, yields non-trivial improvements on various 3D tasks, including semantic segmentation and object detection. Notably, it achieves 73.9% mIoU on the ScanNetv2 semantic segmentation and 72.8% NDS nuScenes object detection benchmarks, ranking 1st on the nuScenes LIDAR leaderboard. It is further boosted to 74.2% NDS with a simple multi-modal fusion. LargeKernel3D attains comparable or superior results than its CNN and transformer counterparts. For the first time, we show that large kernels are feasible and essential for 3D networks.
Pervasive backdoors are triggered by dynamic and pervasive input perturbations. They can be intentionally injected by attackers or naturally exist in normally trained models. They have a different nature from the traditional static and localized backdoors that can be triggered by perturbing a small input area with some fixed pattern, e.g., a patch with solid color. Existing defense techniques are highly effective for traditional backdoors. However, they may not work well for pervasive backdoors, especially regarding backdoor removal and model hardening. In this paper, we propose a novel model hardening technique against pervasive backdoors, including both natural and injected backdoors. We develop a general pervasive attack based on an encoder-decoder architecture enhanced with a special transformation layer. The attack can model a wide range of existing pervasive backdoor attacks and quantify them by class distances. As such, using the samples derived from our attack in adversarial training can harden a model against these backdoor vulnerabilities. Our evaluation on 9 datasets with 15 model structures shows that our technique can enlarge class distances by 59.65% on average with less than 1% accuracy degradation and no robustness loss, outperforming five hardening techniques such as adversarial training, universal adversarial training, MOTH, etc. It can reduce the attack success rate of six pervasive backdoor attacks from 99.06% to 1.94%, surpassing seven state-of-the-art backdoor removal techniques.
In this paper, we propose PETRv2, a unified framework for 3D perception from multi-view images. Based on PETR, PETRv2 explores the effectiveness of temporal modeling, which utilizes the temporal information of previous frames to boost 3D object detection. More specifically, we extend the 3D position embedding (3D PE) in PETR for temporal modeling. The 3D PE achieves the temporal alignment on object position of different frames. A feature-guided position encoder is further introduced to improve the data adaptability of 3D PE. To support for high-quality BEV segmentation, PETRv2 provides a simply yet effective solution by adding a set of segmentation queries. Each segmentation query is responsible for segmenting one specific patch of BEV map. PETRv2 achieves state-of-the-art performance on 3D object detection and BEV segmentation. Detailed robustness analysis is also conducted on PETR framework. We hope PETRv2 can serve as a unified framework for 3D perception.
In this paper, we tackle the problem of learning visual representations from unlabeled scene-centric data. Existing works have demonstrated the potential of utilizing the underlying complex structure within scene-centric data; still, they commonly rely on hand-crafted objectness priors or specialized pretext tasks to build a learning framework, which may harm generalizability. Instead, we propose contrastive learning from data-driven semantic slots, namely SlotCon, for joint semantic grouping and representation learning. The semantic grouping is performed by assigning pixels to a set of learnable prototypes, which can adapt to each sample by attentive pooling over the feature and form new slots. Based on the learned data-dependent slots, a contrastive objective is employed for representation learning, which enhances the discriminability of features, and conversely facilitates grouping semantically coherent pixels together. Compared with previous efforts, by simultaneously optimizing the two coupled objectives of semantic grouping and contrastive learning, our approach bypasses the disadvantages of hand-crafted priors and is able to learn object/group-level representations from scene-centric images. Experiments show our approach effectively decomposes complex scenes into semantic groups for feature learning and significantly benefits downstream tasks, including object detection, instance segmentation, and semantic segmentation. The code will be made publicly available.
The well-designed structures in neural networks reflect the prior knowledge incorporated into the models. However, though different models have various priors, we are used to training them with model-agnostic optimizers (e.g., SGD). In this paper, we propose a novel paradigm of incorporating model-specific prior knowledge into optimizers and using them to train generic (simple) models. As an implementation, we propose a novel methodology to add prior knowledge by modifying the gradients according to a set of model-specific hyper-parameters, which is referred to as Gradient Re-parameterization, and the optimizers are named RepOptimizers. For the extreme simplicity of model structure, we focus on a VGG-style plain model and showcase that such a simple model trained with a RepOptimizer, which is referred to as RepOpt-VGG, performs on par with the recent well-designed models. From a practical perspective, RepOpt-VGG is a favorable base model because of its simple structure, high inference speed and training efficiency. Compared to Structural Re-parameterization, which adds priors into models via constructing extra training-time structures, RepOptimizers require no extra forward/backward computations and solve the problem of quantization. The code and models are publicly available at https://github.com/DingXiaoH/RepOptimizers.
Video captioning is a challenging task as it needs to accurately transform visual understanding into natural language description. To date, state-of-the-art methods inadequately model global-local representation across video frames for caption generation, leaving plenty of room for improvement. In this work, we approach the video captioning task from a new perspective and propose a GL-RG framework for video captioning, namely a \textbf{G}lobal-\textbf{L}ocal \textbf{R}epresentation \textbf{G}ranularity. Our GL-RG demonstrates three advantages over the prior efforts: 1) we explicitly exploit extensive visual representations from different video ranges to improve linguistic expression; 2) we devise a novel global-local encoder to produce rich semantic vocabulary to obtain a descriptive granularity of video contents across frames; 3) we develop an incremental training strategy which organizes model learning in an incremental fashion to incur an optimal captioning behavior. Experimental results on the challenging MSR-VTT and MSVD datasets show that our DL-RG outperforms recent state-of-the-art methods by a significant margin. Code is available at \url{https://github.com/ylqi/GL-RG}.