Accurate detection of mitotic figures in breast histopathology images is central to tumor grading and prognostic assessment. Precise bounding-box annotation remains labor-intensive and variable because mitotic figures are small, morphologically diverse, and boundary-ambiguous. Point-level supervision reduces annotation cost but lacks scale information, making reliable pseudo-box generation essential. Existing point-supervised pipelines often use static or heuristic thresholds that may become unstable as teacher predictions and proposal-score distributions evolve. We propose DDFMitos-Net, a point-supervised teacher-student framework for distribution-aware proposal filtering. The framework learns initial scale priors from point-guided simulated masks, refines teacher-generated pseudo-boxes through Adaptive Multiple Instance Learning, and uses Distribution-based Dynamic Filtering to integrate classification confidence with point-centered spatial information. Adaptive thresholds are estimated with a truncated Dirichlet Process Mixture Model. Transformation-Scale Learning improves geometric consistency, and Center-Aware Domain Adaptation provides auxiliary scanner-aware feature alignment. On MITOS12, MITOS14, TUPAC16, and MIDOG21, DDFMitos-Net achieved repeated-run F1 scores of 0.837 ± 0.004, 0.716 ± 0.006, 0.785 ± 0.005, and 0.806 ± 0.004, respectively. These results indicate stable and competitive point-supervised mitosis detection using low-cost point-level annotations.
Chang Shu, Qiling Tang, Jian-Chi Yue et al.· IEEE journal of biomedical a...· 0 citations
Few-step diffusion models substantially compress temporal computation, making the spatial cost of each model evaluation an increasingly dominant source of inference latency. Progressive-resolution inference reduces this cost by performing early denoising at low resolution and reserving high-resolution computation for refinement. However, existing methods typically lift intermediate latents directly and rely on subsequent steps to absorb the induced distribution mismatch. In the few-step regime, the limited recovery budget leaves these errors as visible artifacts, constraining how late the transition can occur and, consequently, how efficiently it can be performed. We introduce SelfLift, a self-recovering progressive-resolution framework that derives both transition-repair signals and trajectory-aligned supervision from the generative model itself. SelfLift-zero proposes a training-free Artifact-Aware Consistency Lift, using disagreement between direct latent lifting and pixel-VAE re-encoding as both a localized artifact-risk signal and a model-native correction direction. It enables reliable late transitions without external super-resolution, extra denoiser evaluations, or sampling-schedule modifications. Building on this robust transition, SelfLift-rich performs On-Policy Self Recovery on student-visited states, transferring dense high-resolution guidance from an internal self-teacher while remaining aligned with the altered progressive-resolution dynamics. Across FLUX.2-Klein and Z-Image-Turbo, SelfLift reduces end-to-end latency by 41.5% and 44.1%, respectively. Combined with timestep distillation, it delivers overall speedups of 29.61x and 19.21x over the corresponding 50-step models while preserving competitive generation quality, establishing a stronger speed-quality frontier for few-step diffusion.
Ting-Yan Wen, Chen-Qian Yan, Xurui Peng et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
