Research map / Point clouds / Point cloud classification
Neural TMDlayer: Modeling Instantaneous Flow of Features via SDE Generators
Zihang Meng, Vikas Singh, Sathya N. Ravi
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 |
|---|---|---|---|
| use TMDlayer at inference time instead of dropping it (m=1) (ablation) | ResNet-18 + TMDlayer · CIFAR10 (ResNet-18) | accuracy: 75.15 -> 87.35 | Improved |
| increase test batch size m used to build L_m from 1 to 100 (ablation) | ResNet-18 + TMDlayer · CIFAR10 (ResNet-18) | accuracy: 87.35 -> 88.52 | Improved |
| increase test batch size m from 150 to 200 (ablation) | ResNet-18 + TMDlayer · CIFAR10 (ResNet-18) | accuracy: 88.55 -> 88.25 | No clear effect |
| add TMDlayer to ResNet-18 | ResNet-18 · CIFAR10 (ResNet-18) | accuracy: 88.27 -> 88.52 | No clear effect |
| add TMDlayer to ResNet-18 under input noise sigma=0.03 | ResNet-18 · CIFAR10 with input noise | accuracy: 75.85 -> 77.96 | Improved |
| add TMDlayer to ResNet-18 under input noise sigma=0.05 | ResNet-18 · CIFAR10 with input noise | accuracy: 53.87 -> 56.18 | Improved |
| replace PCT transformer layer with TMDlayer on a feed-forward layer | PCT · ModelNet40 classification | overall accuracy: 93.2 -> 93.0 | No clear effect |
| add Chan-Vese update module to the CNN baseline (ablation) | CNN baseline · Vaihingen buildings segmentation | average IoU: 68.9 -> 73.5 | Improved |
| add TMDlayer on top of deep active contour model (ablation) | deep active contour model · Vaihingen buildings segmentation | average IoU: 73.5 -> 74.6 | Improved |
| add TMDlayer to EGNN (non-transductive) | EGNN · miniImageNet 5-way 5-shot | accuracy: 66.85 -> 68.35 | Improved |
| add TMDlayer to EGNN with transduction | EGNN · miniImageNet 5-way 5-shot | accuracy: 76.37 -> 77.78 | Improved |
| add TMDlayer to EGNN-semi with 20% labels | EGNN-semi · miniImageNet semi-supervised | accuracy: 61.88 -> 63.14 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- A-CNN — A-CNN: Annularly Convolutional Neural Networks on Point Clouds (CVPR 2019) · better on 1 of 1 shared result
- PointConv — PointConv: Deep Convolutional Networks on 3D Point Clouds (CVPR 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Neural Stochastic Flows: Solver-Free Modelling and Inference for SDE Solutions (NeurIPS 2025 · similarity 0.87)
- Efficient Training of Neural Stochastic Differential Equations by Matching Finite Dimensional Distributions (ICLR 2025 · similarity 0.86)
- SDE Matching: Scalable and Simulation-Free Training of Latent Stochastic Differential Equations (ICML 2025 · similarity 0.85)
- SING: SDE Inference via Natural Gradients (NeurIPS 2025 · similarity 0.85)
- Variational Inference for SDEs Driven by Fractional Noise (ICLR 2024 · similarity 0.84)
- The Blessing of Randomness: SDE Beats ODE in General Diffusion-based Image Editing (ICLR 2024 · similarity 0.84)
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.
