skip to main content


Title: ISD: Self-Supervised Learning by Iterative Similarity Distillation
Recently, contrastive learning has achieved great results in self-supervised learning, where the main idea is to pull two augmentations of an image (positive pairs) closer compared to other random images (negative pairs). We argue that not all negative images are equally negative. Hence, we introduce a self-supervised learning algorithm where we use a soft similarity for the negative images rather than a binary distinction between positive and negative pairs. We iteratively distill a slowly evolving teacher model to the student model by capturing the similarity of a query image to some random images and transferring that knowledge to the student. Specifically, our method should handle unbalanced and unlabeled data better than existing contrastive learning methods, because the randomly chosen negative set might include many samples that are semantically similar to the query image. In this case, our method labels them as highly similar while standard contrastive methods label them as negatives. Our method achieves comparable results to the state-of-the-art models.  more » « less
Award ID(s):
1845216 1920079 2230693
NSF-PAR ID:
10291163
Author(s) / Creator(s):
; ; ; ;
Date Published:
Journal Name:
International Conference on Computer Vision (ICCV) 2021 1 2020
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Extracting roads in aerial images has numerous applications in artificial intelligence and multimedia computing, including traffic pattern analysis and parking space planning. Learning deep neural networks, though very successful, demands vast amounts of high-quality annotations, of which acquisition is time-consuming and expensive. In this work, we propose a semi-supervised approach for image-based road extraction where only a small set of labeled images are available for training to address this challenge. We design a pixel-wise contrastive loss to self-supervise the network training to utilize the large corpus of unlabeled images. The key idea is to identify pairs of overlapping image regions (positive) or non-overlapping image regions (negative) and encourage the network to make similar outputs for positive pairs or dissimilar outputs for negative pairs. We also develop a negative sampling strategy to filter false negative samples during the process. An iterative procedure is introduced to apply the network over raw images to generate pseudo-labels, filter and select high-quality labels with the proposed contrastive loss, and re-train the network with the enlarged training dataset. We repeat these iterative steps until convergence. We validate the effectiveness of the proposed methods by performing extensive experiments on the public SpaceNet3 and DeepGlobe Road datasets. Results show that our proposed method achieves state-of-the-art results on public image segmentation benchmarks and significantly outperforms other semi-supervised methods.

     
    more » « less
  2. Archaeology has long faced fundamental issues of sampling and scalar representation. Traditionally, the local-to-regional-scale views of settlement patterns are produced through systematic pedestrian surveys. Recently, systematic manual survey of satellite and aerial imagery has enabled continuous distributional views of archaeological phenomena at interregional scales. However, such β€˜brute force’ manual imagery survey methods are both time- and labour-intensive, as well as prone to inter-observer differences in sensitivity and specificity. The development of self-supervised learning methods (e.g. contrastive learning) offers a scalable learning scheme for locating archaeological features using unlabelled satellite and historical aerial images. However, archaeological features are generally only visible in a very small proportion relative to the landscape, while the modern contrastive-supervised learning approach typically yields an inferior performance on highly imbalanced datasets. In this work, we propose a framework to address this long-tail problem. As opposed to the existing contrastive learning approaches that typically treat the labelled and unlabelled data separately, our proposed method reforms the learning paradigm under a semi-supervised setting in order to fully utilize the precious annotated data (<7% in our setting). Specifically, the highly unbalanced nature of the data is employed as the prior knowledge in order to form pseudo negative pairs by ranking the similarities between unannotated image patches and annotated anchor images. In this study, we used 95,358 unlabelled images and 5,830 labelled images in order to solve the issues associated with detecting ancient buildings from a long-tailed satellite image dataset. From the results, our semi-supervised contrastive learning model achieved a promising testing balanced accuracy of 79.0%, which is a 3.8% improvement as compared to other state-of-the-art approaches. 
    more » « less
  3. Unsupervised learning has recently made exceptional progress because of the development of more effective contrastive learning methods. However, CNNs are prone to depend on low-level features that humans deem non-semantic. This dependency has been conjectured to induce a lack of robustness to image perturbations or domain shift. In this paper, we show that by generating carefully designed negative samples, contrastive learning can learn more robust representations with less dependence on such features. Contrastive learning utilizes positive pairs which preserve semantic information while perturbing superficial features in the training images. Similarly, we propose to generate negative samples in a reversed way, where only the superfluous instead of the semantic features are preserved. We develop two methods, texture-based and patch-based augmentations, to generate negative samples. These samples achieve better generalization, especially under out-of-domain settings. We also analyze our method and the generated texture-based samples, showing that texture features are indispensable in classifying particular ImageNet classes and especially finer classes. We also show that the model bias between texture and shape features favors them differently under different test settings. 
    more » « less
  4. Contrastive learning (CL), a self-supervised learning approach, can effectively learn visual representations from unlabeled data. Given the CL training data, generative models can be trained to generate synthetic data to supplement the real data. Using both synthetic and real data for CL training has the potential to improve the quality of learned representations. However, synthetic data usually has lower quality than real data, and using synthetic data may not improve CL compared with using real data. To tackle this problem, we propose a data generation framework with two methods to improve CL training by joint sample generation and contrastive learning. The first approach generates hard samples for the main model. The generator is jointly learned with the main model to dynamically customize hard samples based on the training state of the main model. Besides, a pair of data generators are proposed to generate similar but distinct samples as positive pairs. In joint learning, the hardness of a positive pair is progressively increased by decreasing their similarity. Experimental results on multiple datasets show superior accuracy and data efficiency of the proposed data generation methods applied to CL. For example, about 4.0%, 3.5%, and 2.6% accuracy improvements for linear classification are observed on ImageNet-100, CIFAR-100, and CIFAR-10, respectively. Besides, up to 2Γ— data efficiency for linear classification and up to 5Γ— data efficiency for transfer learning are achieved. 
    more » « less
  5. Graph representation learning is crucial for many real-world ap- plications (e.g. social relation analysis). A fundamental problem for graph representation learning is how to effectively learn rep- resentations without human labeling, which is usually costly and time-consuming. Graph contrastive learning (GCL) addresses this problem by pulling the positive node pairs (or similar nodes) closer while pushing the negative node pairs (or dissimilar nodes) apart in the representation space. Despite the success of the existing GCL methods, they primarily sample node pairs based on the node- level proximity yet the community structures have rarely been taken into consideration. As a result, two nodes from the same community might be sampled as a negative pair. We argue that the community information should be considered to identify node pairs in the same communities, where the nodes insides are seman- tically similar. To address this issue, we propose a novel Graph Communal Contrastive Learning (π‘”πΆπ‘œπ‘œπΏ) framework to jointly learn the community partition and learn node representations in an end-to-end fashion. Specifically, the proposed π‘”πΆπ‘œπ‘œπΏ consists of two components: a Dense Community Aggregation (𝐷𝑒𝐢𝐴) algo- rithm for community detection and a Reweighted Self-supervised Cross-contrastive (𝑅𝑒𝑆𝐢) training scheme to utilize the community information. Additionally, the real-world graphs are complex and often consist of multiple views. In this paper, we demonstrate that the proposed π‘”πΆπ‘œπ‘œπΏ can also be naturally adapted to multiplex graphs. Finally, we comprehensively evaluate the proposed π‘”πΆπ‘œπ‘œπΏ on a variety of real-world graphs. The experimental results show that the π‘”πΆπ‘œπ‘œπΏ outperforms the state-of-the-art methods. 
    more » « less