2022-04-01 更新
Long-Short Temporal Contrastive Learning of Video Transformers
Authors:Jue Wang, Gedas Bertasius, Du Tran, Lorenzo Torresani
Video transformers have recently emerged as a competitive alternative to 3D CNNs for video understanding. However, due to their large number of parameters and reduced inductive biases, these models require supervised pretraining on large-scale image datasets to achieve top performance. In this paper, we empirically demonstrate that self-supervised pretraining of video transformers on video-only datasets can lead to action recognition results that are on par or better than those obtained with supervised pretraining on large-scale image datasets, even massive ones such as ImageNet-21K. Since transformer-based models are effective at capturing dependencies over extended temporal spans, we propose a simple learning procedure that forces the model to match a long-term view to a short-term view of the same video. Our approach, named Long-Short Temporal Contrastive Learning (LSTCL), enables video transformers to learn an effective clip-level representation by predicting temporal context captured from a longer temporal extent. To demonstrate the generality of our findings, we implement and validate our approach under three different self-supervised contrastive learning frameworks (MoCo v3, BYOL, SimSiam) using two distinct video-transformer architectures, including an improved variant of the Swin Transformer augmented with space-time attention. We conduct a thorough ablation study and show that LSTCL achieves competitive performance on multiple video benchmarks and represents a convincing alternative to supervised image-based pretraining.
PDF Accepted in CVPR 2022
论文截图
Semantic Pose Verification for Outdoor Visual Localization with Self-supervised Contrastive Learning
Authors:Semih Orhan, Jose J. Guerrero, Yalin Bastanlar
Any city-scale visual localization system has to overcome long-term appearance changes, such as varying illumination conditions or seasonal changes between query and database images. Since semantic content is more robust to such changes, we exploit semantic information to improve visual localization. In our scenario, the database consists of gnomonic views generated from panoramic images (e.g. Google Street View) and query images are collected with a standard field-of-view camera at a different time. To improve localization, we check the semantic similarity between query and database images, which is not trivial since the position and viewpoint of the cameras do not exactly match. To learn similarity, we propose training a CNN in a self-supervised fashion with contrastive learning on a dataset of semantically segmented images. With experiments we showed that this semantic similarity estimation approach works better than measuring the similarity at pixel-level. Finally, we used the semantic similarity scores to verify the retrievals obtained by a state-of-the-art visual localization method and observed that contrastive learning-based pose verification increases top-1 recall value to 0.90 which corresponds to a 2% improvement.
PDF
论文截图