Match me if you can: Semantic Correspondence Learning with Unpaired Images

Recent approaches for semantic correspondence have focused on obtaining high-quality correspondences using a complicated network, refining the ambiguous or noisy matching points. Despite their performance improvements, they remain constrained by the limited training pairs due to costly point-level annotations. This paper proposes a simple yet effective method that performs training with unlabeled pairs to complement both limited image pairs and sparse point pairs, requiring neither extra labeled keypoints nor trainable modules. We fundamentally extend the data quantity and variety by augmenting new unannotated pairs not primitively provided as training pairs in benchmarks. Using a simple teacher-student framework, we offer reliable pseudo correspondences to the student network via machine supervision. Finally, the performance of our network is steadily improved by the proposed iterative training, putting back the student as a teacher to generate refined labels and train a new student repeatedly. Our models outperform the milestone baselines, including state-of-the-art methods on semantic correspondence benchmarks.

Longer-range Contextualized Masked Autoencoder

Masked image modeling (MIM) has emerged as a promising self-supervised learning (SSL) strategy. The MIM pre-training facilitates learning powerful representations using an encoder-decoder framework by randomly masking some input pixels and reconstructing the masked pixels from the remaining ones. However, as the encoder is trained with partial pixels, the MIM pre-training can suffer from a low capability of understanding long-range dependency. This limitation may hinder its capability to fully understand multiple-range dependencies, resulting in narrow highlighted regions in the attention map that may incur accuracy drops. To mitigate the limitation, We propose a self-supervised learning framework, named Longer-range Contextualized Masked Autoencoder (LC-MAE). LC-MAE effectively leverages a global context understanding of visual representations while simultaneously reducing the spatial redundancy of input at the same time. Our method steers the encoder to learn from entire pixels in multiple views while also learning local representation from sparse pixels. As a result, LC-MAE learns more discriminative representations, leading to a performance improvement of achieving 84.2% top-1 accuracy with ViT-B on ImageNet-1K with 0.6%p gain. We attribute the success to the enhanced pre-training method, as evidenced by the singular value spectrum and attention analyses. Finally, LC-MAE achieves significant performance gains at the downstream semantic segmentation and fine-grained visual classification tasks; and on diverse robust evaluation metrics. Our code will be publicly available.

Augmenting Sub-model to Improve Main Model

Image classification has improved with the development of training techniques. However, these techniques often require careful parameter tuning to balance the strength of regularization, limiting their potential benefits. In this paper, we propose a novel way to use regularization called Augmenting Sub-model (AugSub). AugSub consists of two models: the main model and the sub-model. While the main model employs conventional training recipes, the sub-model leverages the benefit of additional regularization. AugSub achieves this by mitigating adverse effects through a relaxed loss function similar to self-distillation loss. We demonstrate the effectiveness of AugSub with three drop techniques: dropout, drop-path, and random masking. Our analysis shows that all AugSub improves performance, with the training loss converging even faster than regular training. Among the three, AugMask is identified as the most practical method due to its performance and cost efficiency. We further validate AugMask across diverse training recipes, including DeiT-III, ResNet, MAE fine-tuning, and Swin Transformer. The results show that AugMask consistently provides significant performance gain. AugSub provides a practical and effective solution for introducing additional regularization under various training recipes. Code is available at \url{https://github.com/naver-ai/augsub}.

What Do Self-Supervised Vision Transformers Learn?

We present a comparative study on how and why contrastive learning (CL) and masked image modeling (MIM) differ in their representations and in their performance of downstream tasks. In particular, we demonstrate that self-supervised Vision Transformers (ViTs) have the following properties: (1) CL trains self-attentions to capture longer-range global patterns than MIM, such as the shape of an object, especially in the later layers of the ViT architecture. This CL property helps ViTs linearly separate images in their representation spaces. However, it also makes the self-attentions collapse into homogeneity for all query tokens and heads. Such homogeneity of self-attention reduces the diversity of representations, worsening scalability and dense prediction performance. (2) CL utilizes the low-frequency signals of the representations, but MIM utilizes high-frequencies. Since low- and high-frequency information respectively represent shapes and textures, CL is more shape-oriented and MIM more texture-oriented. (3) CL plays a crucial role in the later layers, while MIM mainly focuses on the early layers. Upon these analyses, we find that CL and MIM can complement each other and observe that even the simplest harmonization can help leverage the advantages of both methods. The code is available at https://github.com/naver-ai/cl-vs-mim.

SeiT: Storage-Efficient Vision Training with Tokens Using 1% of Pixel Storage

We need billion-scale images to achieve more generalizable and ground-breaking vision models, as well as massive dataset storage to ship the images (e.g., the LAION-4B dataset needs 240TB storage space). However, it has become challenging to deal with unlimited dataset storage with limited storage infrastructure. A number of storage-efficient training methods have been proposed to tackle the problem, but they are rarely scalable or suffer from severe damage to performance. In this paper, we propose a storage-efficient training strategy for vision classifiers for large-scale datasets (e.g., ImageNet) that only uses 1024 tokens per instance without using the raw level pixels; our token storage only needs <1% of the original JPEG-compressed raw pixels. We also propose token augmentations and a Stem-adaptor module to make our approach able to use the same architecture as pixel-based approaches with only minimal modifications on the stem layer and the carefully tuned optimization settings. Our experimental results on ImageNet-1k show that our method significantly outperforms other storage-efficient training methods with a large gap. We further show the effectiveness of our method in other practical scenarios, storage-efficient pre-training, and continual learning. Code is available at https://github.com/naver-ai/seit

Group Generalized Mean Pooling for Vision Transformer

Vision Transformer (ViT) extracts the final representation from either class token or an average of all patch tokens, following the architecture of Transformer in Natural Language Processing (NLP) or Convolutional Neural Networks (CNNs) in computer vision. However, studies for the best way of aggregating the patch tokens are still limited to average pooling, while widely-used pooling strategies, such as max and GeM pooling, can be considered. Despite their effectiveness, the existing pooling strategies do not consider the architecture of ViT and the channel-wise difference in the activation maps, aggregating the crucial and trivial channels with the same importance. In this paper, we present Group Generalized Mean (GGeM) pooling as a simple yet powerful pooling strategy for ViT. GGeM divides the channels into groups and computes GeM pooling with a shared pooling parameter per group. As ViT groups the channels via a multi-head attention mechanism, grouping the channels by GGeM leads to lower head-wise dependence while amplifying important channels on the activation maps. Exploiting GGeM shows 0.1%p to 0.7%p performance boosts compared to the baselines and achieves state-of-the-art performance for ViT-Base and ViT-Large models in ImageNet-1K classification task. Moreover, GGeM outperforms the existing pooling strategies on image retrieval and multi-modal representation learning tasks, demonstrating the superiority of GGeM for a variety of tasks. GGeM is a simple algorithm in that only a few lines of code are necessary for implementation.

Similarity of Neural Architectures Based on Input Gradient Transferability

In this paper, we aim to design a quantitative similarity function between two neural architectures. Specifically, we define a model similarity using input gradient transferability. We generate adversarial samples of two networks and measure the average accuracy of the networks on adversarial samples of each other. If two networks are highly correlated, then the attack transferability will be high, resulting in high similarity. Using the similarity score, we investigate two topics: (1) Which network component contributes to the model diversity? (2) How does model diversity affect practical scenarios? We answer the first question by providing feature importance analysis and clustering analysis. The second question is validated by two different scenarios: model ensemble and knowledge distillation. Our findings show that model diversity takes a key role when interacting with different neural architectures. For example, we found that more diversity leads to better ensemble performance. We also observe that the relationship between teacher and student networks and distillation performance depends on the choice of the base architecture of the teacher and student networks. We expect our analysis tool helps a high-level understanding of differences between various neural architectures as well as practical guidance when using multiple architectures.

Scratching Visual Transformer's Back with Uniform Attention

The favorable performance of Vision Transformers (ViTs) is often attributed to the multi-head self-attention (MSA). The MSA enables global interactions at each layer of a ViT model, which is a contrasting feature against Convolutional Neural Networks (CNNs) that gradually increase the range of interaction across multiple layers. We study the role of the density of the attention. Our preliminary analyses suggest that the spatial interactions of attention maps are close to dense interactions rather than sparse ones. This is a curious phenomenon, as dense attention maps are harder for the model to learn due to steeper softmax gradients around them. We interpret this as a strong preference for ViT models to include dense interaction. We thus manually insert the uniform attention to each layer of ViT models to supply the much needed dense interactions. We call this method Context Broadcasting, CB. We observe that the inclusion of CB reduces the degree of density in the original attention maps and increases both the capacity and generalizability of the ViT models. CB incurs negligible costs: 1 line in your model code, no additional parameters, and minimal extra operations.

Improving Ensemble Distillation With Weight Averaging and Diversifying Perturbation

Ensembles of deep neural networks have demonstrated superior performance, but their heavy computational cost hinders applying them for resource-limited environments. It motivates distilling knowledge from the ensemble teacher into a smaller student network, and there are two important design choices for this ensemble distillation: 1) how to construct the student network, and 2) what data should be shown during training. In this paper, we propose a weight averaging technique where a student with multiple subnetworks is trained to absorb the functional diversity of ensemble teachers, but then those subnetworks are properly averaged for inference, giving a single student network with no additional inference cost. We also propose a perturbation strategy that seeks inputs from which the diversities of teachers can be better transferred to the student. Combining these two, our method significantly improves upon previous methods on various image classification tasks.

An Extendable, Efficient and Effective Transformer-based Object Detector

Transformers have been widely used in numerous vision problems especially for visual recognition and detection. Detection transformers are the first fully end-to-end learning systems for object detection, while vision transformers are the first fully transformer-based architecture for image classification. In this paper, we integrate Vision and Detection Transformers (ViDT) to construct an effective and efficient object detector. ViDT introduces a reconfigured attention module to extend the recent Swin Transformer to be a standalone object detector, followed by a computationally efficient transformer decoder that exploits multi-scale features and auxiliary techniques essential to boost the detection performance without much increase in computational load. In addition, we extend it to ViDT+ to support joint-task learning for object detection and instance segmentation. Specifically, we attach an efficient multi-scale feature fusion layer and utilize two more auxiliary training losses, IoU-aware loss and token labeling loss. Extensive evaluation results on the Microsoft COCO benchmark dataset demonstrate that ViDT obtains the best AP and latency trade-off among existing fully transformer-based object detectors, and its extended ViDT+ achieves 53.2AP owing to its high scalability for large models. The source code and trained models are available at https://github.com/naver-ai/vidt.