Research map / Learning theory and algorithms / Optimal transport
Optimal transport unlocks end-to-end learning for single-molecule localization
Romain Seailles, Jean Baptiste Masson, Jean Ponce, Julien Mairal
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add optimal-transport loss to DECODE architecture (ablation) | DECODE · EPFL 2016 SMLM (density 2.0, high SNR) | E3D: 0.705 -> 0.740 | Improved |
| add iterative refinement architecture on top of OT loss (ablation) | OT-loss model · EPFL 2016 SMLM (density 2.0, high SNR) | E3D: 0.740 -> 0.750 | Improved |
| add iterative architecture with DECODE loss (ablation) | DECODE loss · EPFL 2016 SMLM (density 2.0, high SNR) | E3D: 0.705 -> 0.703 | No clear effect |
| use CNN decoder instead of ViT decoder (ablation) | Ours (d=HW/4) · MT0N1HDAS | E3D: 0.722 -> 0.750 | Improved |
| reduce candidates from HW to HW/4 (ablation) | CNN decoder · MT0N1HDAS | E3D: 0.732 -> 0.750 | Improved |
| reduce candidates from HW/4 to HW/16 (ablation) | CNN decoder · MT0N1HDAS | E3D: 0.750 -> 0.739 | Hurt |
| increase Sinkhorn epsilon from 1e-4 to 1e-2 (ablation) | OT loss · MT0N1HDAS | E3D: 0.750 -> 0.397 | Hurt |
| lower epsilon from 1e-3 to 1e-4 (ablation) | OT loss · MT0N1HDAS | E3D: 0.739 -> 0.750 | Improved |
| lower epsilon from 1e-4 to 1e-5 (ablation) | OT loss · MT0N1HDAS | E3D: 0.750 -> 0.747 | No clear effect |
| use Sinkhorn regularized OT instead of Hungarian matching (ablation) | OT loss · MT0N1HDAS | E3D: 0.742 -> 0.750 | Improved |
| increase camera parameter jitter from sigma 0 to 0.30 at test time (ablation) | Ours · EPFL synthetic density 2.0 (camera mismatch) | E3D: 0.811 -> 0.805 | No clear effect |
| use OT loss and iterative arch vs LiteLoc under binning x32 | Ours vs LiteLoc · Real NPC-Nup96 x32 | FRC (nm): 71.5 -> 44.2 | Improved |
Closest papers by meaning
- Masked Autoencoders for Microscopy are Scalable Learners of Cellular Biology (CVPR 2024 · similarity 0.83)
- ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy (ICML 2025 · similarity 0.83)
- scSplit: Bringing Severity Cognizance to Image Decomposition in Fluorescence Microscopy (NeurIPS 2025 · similarity 0.83)
- Micro-Bench: A Microscopy Benchmark for Vision-Language Understanding (NeurIPS 2024 · similarity 0.83)
- Langevin-Informed Transfer Learning: Replacing the Target Samples by Black-Box Feedback (NeurIPS 2026 · similarity 0.82)
- Unbalancedness in Neural Monge Maps Improves Unpaired Domain Translation (ICLR 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
