The paper reports a controlled study of self‑supervised learning (SSL) objectives for image and video pretraining under limited data, architecture, and compute budgets. It compares contrastive, reconstruction, feature‑prediction, and diffusion methods, finding that DINOv2‑style pretraining delivers the best overall performance. Combining DINOv2 with video SSL objectives such as VideoMAE improves image classification and segmentation but harms video tracking and camera‑pose estimation, highlighting a trade‑off between semantic and geometric learning.
Self-supervision is a powerful technique for learning visual representations from unlabeled data. Existing techniques primarily adopt a two-stage approach for self-supervised learning (SSL): a pretraining stage on unlabeled data followed by a finetuning stage on labeled data.
The paper introduces a biologically inspired framework that learns object‑centric visual representations from raw videos without human annotations or camera calibration. By using motion boundaries detected via optical flow and clustering to create pseudo‑instance masks, the method supervises a single‑image encoder with pixel‑level pairwise metric learning. Training on 195 million pseudo‑labeled frames and expanding to 421 million frames through Motion‑Verified Self‑Training, the approach yields Swin‑based encoders that outperform or match supervised and self‑supervised baselines on tasks such as monocular depth estimation, 3D object detection, 3D occupancy prediction, and end‑to‑end planning.
LeVJEPA is a video encoder that eliminates the need for architectural asymmetries, exponential-moving-average target encoders, stop-gradients, and capacity-limited predictors used in prior self‑supervised methods. It trains a single encoder with an invariance loss over global and local views, regularized by SIGReg to prevent collapse, and achieves strong performance with far less pretraining compute. The approach also allows block‑causal attention, making temporal ordering a property of the encoder itself, and matches or surpasses state‑of‑the‑art baselines on both appearance‑centric and motion‑centric benchmarks.
The paper investigates the effect of using semantic positive pairs—different instances of the same class—in self‑supervised visual representation learning. By creating matched ImageNet‑1K subsets of augmented pairs and manually curated semantic pairs, the authors compare contrastive and non‑contrastive SSL methods under identical training conditions. Across transfer learning and object detection tasks, semantic‑pair pretraining consistently outperforms augmented‑pair pretraining, with contrastive methods like SimCLR showing the largest gains, indicating that semantic pairs foster additional invariances beyond standard augmentations.
arXiv:2609.40347v1 Announce Type: new Abstract: We introduce VideoMSN, a Masked Siamese Network framework for efficient self-supervised spatio-temporal representation learning in videos. Instead of r...
arXiv:2608. 03218v1 Announce Type: cross Abstract: Dataset distillation compresses a large training set into a compact synthetic set while retaining its downstream utility.
Dataset distillation compresses a large training set into a compact synthetic set while retaining its downstream utility. Most existing methods target randomly initialized networks, whereas modern vision systems often adapt frozen pretrained encoders with lightweight modules.
arXiv:2407.03463v2 Announce Type: replace-cross Abstract: In the realm of self-supervised learning (SSL), conventional wisdom has gravitated towards the utility of massive, general domain datasets fo...
The paper introduces GeoNeXt, a framework that repurposes pretrained video generative models for geometry estimation by framing it as a next‑frame prediction task. Unlike prior methods that either train separate depth/normal models or fine‑tune image diffusion backbones, GeoNeXt jointly models images and geometric targets, leveraging the structured knowledge of video models for more data‑efficient learning. Experiments show zero‑shot monocular depth and surface normal estimation that outperforms existing generative approaches and rivals discriminative state‑of‑the‑art methods while using far less training data.
The paper investigates whether pretrained image models can generalize to unseen datasets by clustering their embeddings. Using encoders trained only on ImageNet‑1k, both supervised and self‑supervised, the authors evaluate clustering performance on out‑of‑domain images. They find that supervised encoders perform better within the training domain, while self‑supervised encoders excel far outside it, and that fine‑tuning self‑supervised models reverses this trend. Additionally, the study shows that the silhouette score in UMAP‑reduced space correlates strongly with clustering accuracy, offering a proxy metric when labels are unavailable.
arXiv:2602. 02381v2 Announce Type: replace Abstract: Joint-embedding self-supervised learning (SSL), the key paradigm for unsupervised representation learning from visual data, learns from invariances between semantically-related data pairs.
arXiv:2607. 00784v1 Announce Type: cross Abstract: Vision-language pretraining remains dominated by contrastive objectives, whereas vision-only self-supervised learning has largely adopted non-contrastive methods.