19 October 2023 in IEEE Access
Self-supervised learning (SSL) has emerged as a promising approach for learning representations from unlabeled data. Momentum-based contrastive frameworks such as MoCo-v3 have shown remarkable success among the many SSL methods proposed in recent years. However, a significant gap in encoder representation exists between the online encoder (student) and the momentum encoder (teacher) in these frameworks, limiting the performance on downstream tasks. We identify this gap as a bottleneck often overlooked in existing frameworks and propose “residual momentum” that explicitly reduces the gap during training to encourage the student to learn representations closer to the teacher’s. We also reveal that a similar technique, knowledge distillation (KD), to reduce the distribution gap with cross-entropy-based loss in supervised learning is useless in the SSL context and demonstrate that the intra-representation gap measured by cosine similarity is crucial for EMA-based SSLs. Extensive experiments on different benchmark datasets and architectures demonstrate the superiority of our method compared to state-of-the-art contrastive learning baselines. Specifically, our method outperforms MoCo-v3 0.7% top-1 in ImageNet, 2.82% on CIFAR-100, 1.8% AP, and 3.0% AP75 on VOC detection pre-trained on the COCO dataset; it also improves DenseCL with 0.5% AP (800ep) and 0.6% AP75 (1600ep). Our work highlights the importance of reducing the teacher-student intra-gap in momentum-based contrastive learning frameworks and provides a practical solution for improving the quality of learned representations.
21 November 2022 in arXiv
Self-supervised learning (SSL) approaches have shown promising capabilities in learning the representation from unlabeled data. Amongst them, momentum-based frameworks have attracted significant attention. Despite being a great success, these momentum-based SSL frameworks suffer from a large gap in representation between the online encoder (student) and the momentum encoder (teacher), which hinders performance on downstream tasks. This paper is the first to investigate and identify this invisible gap as a bottleneck that has been overlooked in the existing SSL frameworks, potentially preventing the models from learning good representation. To solve this problem, we propose "residual momentum" to directly reduce this gap to encourage the student to learn the representation as close to that of the teacher as possible, narrow the performance gap with the teacher, and significantly improve the existing SSL. Our method is straightforward, easy to implement, and can be easily plugged into other SSL frameworks. Extensive experimental results on numerous benchmark datasets and diverse network architectures have demonstrated the effectiveness of our method over the state-of-the-art contrastive learning baselines.
13 November 2024 in arXiv
Diffusion models have recently emerged as a potent tool in generative modeling. However, their inherent iterative nature often results in sluggish image generation due to the requirement for multiple model evaluations. Recent progress has unveiled the intrinsic link between diffusion models and Probability Flow Ordinary Differential Equations (ODEs), thus enabling us to conceptualize diffusion models as ODE systems. Simultaneously, Physics Informed Neural Networks (PINNs) have substantiated their effectiveness in solving intricate differential equations through implicit modeling of their solutions. Building upon these foundational insights, we introduce Physics Informed Distillation (PID), which employs a student model to represent the solution of the ODE system corresponding to the teacher diffusion model, akin to the principles employed in PINNs. Through experiments on CIFAR 10 and ImageNet 64x64, we observe that PID achieves performance comparable to recent distillation methods. Notably, it demonstrates predictable trends concerning method-specific hyperparameters and eliminates the need for synthetic dataset generation during the distillation process. Both of which contribute to its easy-to-use nature as a distillation approach for Diffusion Models. Our code and pre-trained checkpoint are publicly available at: https://github.com/pantheon5100/pid_diffusion.git.
30 November 2023 in arXiv
It is well known the adversarial optimization of GAN-based image super-resolution (SR) methods makes the preceding SR model generate unpleasant and undesirable artifacts, leading to large distortion. We attribute the cause of such distortions to the poor calibration of the discriminator, which hampers its ability to provide meaningful feedback to the generator for learning high-quality images. To address this problem, we propose a simple but non-travel diffusion-style data augmentation scheme for current GAN-based SR methods, known as DifAugGAN. It involves adapting the diffusion process in generative diffusion models for improving the calibration of the discriminator during training motivated by the successes of data augmentation schemes in the field to achieve good calibration. Our DifAugGAN can be a Plug-and-Play strategy for current GAN-based SISR methods to improve the calibration of the discriminator and thus improve SR performance. Extensive experimental evaluations demonstrate the superiority of DifAugGAN over state-of-the-art GAN-based SISR methods across both synthetic and real-world datasets, showcasing notable advancements in both qualitative and quantitative results.