We study a practical domain adaptation task, called source-free unsupervised domain adaptation (UDA) problem, in which we cannot access source domain data due to data privacy issues but only a pre-trained source model and unlabeled target data are available. This task, however, is very difficult due to one key challenge: the lack of source data and target domain labels makes model adaptation very challenging. To address this, we propose to mine the hidden knowledge in the source model and exploit it to generate source avatar prototypes (i.e. representative features for each source class) as well as target pseudo labels for domain alignment. To this end, we propose a Contrastive Prototype Generation and Adaptation (CPGA) method. Specifically, CPGA consists of two stages: (1) prototype generation: by exploring the classification boundary information of the source model, we train a prototype generator to generate avatar prototypes via contrastive learning. (2) prototype adaptation: based on the generated source prototypes and target pseudo labels, we develop a new robust contrastive prototype adaptation strategy to align each pseudo-labeled target data to the corresponding source prototypes. Extensive experiments on three UDA benchmark datasets demonstrate the effectiveness and superiority of the proposed method.

We study a practical domain adaptation task, called source-free unsupervised domain adaptation (UDA) problem, in which we cannot access source domain data due to data privacy issues but only a pre-trained source model and unlabeled target data are available. This task, however, is very difficult due to one key challenge: the lack of source data and target domain labels makes model adaptation very challenging. To address this, we propose to mine the hidden knowledge in the source model and exploit it to generate source avatar prototypes (i.e., representative features for each source class) as well as target pseudo labels for domain alignment. To this end, we propose a Contrastive Prototype Generation and Adaptation (CPGA) method. Specifically, CPGA consists of two stages: (1) prototype generation: by exploring the classification boundary information of the source model, we train a prototype generator to generate avatar prototypes via contrastive learning. (2) prototype adaptation: based on the generated source prototypes and target pseudo labels, we develop a new robust contrastive prototype adaptation strategy to align each pseudo-labeled target data to the corresponding source prototypes. Extensive experiments on three UDA benchmark datasets demonstrate the effectiveness and superiority of the proposed method.

No abstract

No abstract

Background To evaluate the ability of optical coherence tomography angiography (OCTA) to quantitatively assess the effectiveness of intra‐arterial thrombolysis (IAT) surgery for the treatment of retinal artery occlusion (RAO). Methods This was a prospective observational study. Patients diagnosed with RAO who received IAT were enrolled. All participants underwent comprehensive ophthalmologic examinations and OCTA scans. The best‐corrected visual acuity (BCVA), vascular densities (VDs) of the superficial capillary plexus (SCP), deep capillary plexus (DCP), radial peripapillary capillaries, and central retinal thickness (CRT) were recorded. The above parameters were compared between RAO eyes and the fellow healthy eyes before and after IAT. Correlations between the BCVA and OCTA parameters in RAO eyes were calculated. Results Thirty‐four eyes from 34 consecutive RAO patients with a mean age of 51.0 ± 12.9 years were enrolled. There was a considerable decrease of VDs in the SCP, DCP, and radial peripapillary capillaries in all RAO eyes (all p < 0.001). Compared with contralateral normal eyes, CRT was significantly increased in RAO eyes (p < 0.001). The SCP and DCP VDs were significantly improved after IAT surgery (p = 0.010 and 0.014, respectively). BCVA in logMAR unit was negatively correlated with the DCP VD (r = −0.664, p = 0.010) and positively correlated with CRT (r = 0.597, p = 0.024) after surgery, but not significantly correlated with VDs or CRT before surgery. Conclusion Macular and peripapillary VDs of the retina detected by OCTA were greatly decreased in RAO eyes and improved after IAT treatment. OCTA is capable of quantifying VDs in separate retinal layers non‐invasively, conveniently, efficiently, and precisely.

No abstract

No abstract