Research map / Probabilistic and scientific modeling / Molecules and proteins
AIM: An Auto-Augmenter for Images and Meshes
Vinit Veerendraveer Singh, Chandra Kambhamettu
Check 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 sample-aware AIM deformations | Image classifiers · CUB-200 and Oxford-IIT Pets | accuracy: 78.3 -> 79.4; 91.6 -> 91.9 (ResNet-18); 83.5 -> 84.0; 93.7 -> 93.7 (EfficientNet-b2) | Mixed |
| Add AIM deformations | MeshNet · MSB and SHREC-11 split-16 | accuracy: 56.5 -> 67.3; 55.6 -> 63.1 | Improved |
| Add AIM deformations | MeshNet++ · MSB and SHREC-11 split-16 | accuracy: 94.5 -> 95.6; 100 -> 100 | Mixed |
| Add AIM deformations | DiffusionNet · Human body mesh segmentation | accuracy: 90.9 -> 91.3 | Improved |
| Replace learned AIM deformations with random vertex jitter | DiffusionNet · Human body mesh segmentation | accuracy: 90.9 -> 87.9 | Hurt |
| Add directional consistency loss (ablation) | AIM image classifiers · CUB-200 and Oxford-IIT Pets | accuracy: ResNet-18: 78.7 -> 79.4, 91.3 -> 91.9; ResNet-34: 80.4 -> 79.8, 92.6 -> 93 | Mixed |
| Add directional consistency loss (ablation) | AIM image classifiers · CUB-200 and Oxford-IIT Pets | accuracy: ResNet-50: 82.3 -> 82.5, 93.5 -> 93.5; EfficientNet-b0: 81.8 -> 82.8, 93.2 -> 93.4 | Mixed |
| Add directional consistency loss (ablation) | AIM image classifiers · CUB-200 and Oxford-IIT Pets | accuracy: EfficientNet-b1: 83.1 -> 83.1, 93.2 -> 93.0; EfficientNet-b2: 83.3 -> 84, 93.2 -> 93.7 | Mixed |
| Use AIM deformations during testing as well as training (ablation) | Image classifiers · CUB-200 and Oxford-IIT Pets | accuracy: ResNet-18: 78.8 -> 79.4, 91.7 -> 91.9; ResNet-34: 80.1 -> 80.4, 92.8 -> 93.0; ResNet-50: 82.0 -> 82.5, 93.3 -> 93.5 | Mixed |
Closest papers by meaning
- Mesh-Based Autoencoders for Localized Deformation Component Analysis (AAAI 2018 · similarity 0.87)
- Faster AutoAugment: Learning Augmentation Strategies Using Backpropagation (ECCV 2020 · similarity 0.87)
- Fully Convolutional Mesh Autoencoder using Efficient Spatially Varying Kernels (NeurIPS 2020 · similarity 0.87)
- Adversarial Data Augmentation via Deformation Statistics (ECCV 2020 · similarity 0.87)
- Variational Autoencoders for Deforming 3D Mesh Models (CVPR 2018 · similarity 0.87)
- Data-Efficient Augmentation for Training Neural Networks (NeurIPS 2022 · similarity 0.86)
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.
