Research map / Architectures and efficiency / CNN design
Multi-Scale Dense Networks for Resource Efficient Image Classification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 dense connectivity between layers and classifiers (ablation) | MSDNet · CIFAR-100 ablation | accuracy across budgets: drastic accuracy drop when removed | Helped |
| add multi-scale feature maps (ablation) | MSDNet · CIFAR-100 ablation | accuracy at low budgets: removal hurts only in lower budget regions | Helped |
| add intermediate classifiers with dense connectivity and multi-scale features | ResNet/DenseNet · CIFAR-100 anytime | accuracy | Helped |
| multi-scale dense architecture vs. ensembles of ResNets/DenseNets | anytime prediction on ImageNet · ImageNet anytime | top-1 accuracy: ~4%-8% higher at 0.1-0.3x10^10 FLOPs | Helped |
| ensembles vs MSDNet at very small budgets | anytime prediction on CIFAR-100 · CIFAR-100 anytime | accuracy at very low budget | Mixed |
| dynamic evaluation (early exit of easy examples) in budgeted batch setting | MSDNet · ImageNet budgeted batch | top-1 accuracy at fixed average budget: ~6% higher than ResNet at 1.7x10^9 FLOPs | Helped |
| double growth rate after each transition layer (DenseNet*) | DenseNet · CIFAR-100 anytime and budgeted batch | accuracy vs computation | Helped |
Closest papers by meaning
- Densely Connected Convolutional Networks (CVPR 2017 · similarity 0.89)
- Glance and Focus: a Dynamic Approach to Reducing Spatial Redundancy in Image Classification (NeurIPS 2020 · similarity 0.88)
- Contextual Convolutional Networks (ICLR 2023 · similarity 0.88)
- ImageNet-Hard: The Hardest Images Remaining from a Study of the Power of Zoom and Spatial Biases in Image Classification (NeurIPS 2023 · similarity 0.87)
- Accelerated Training for Massive Classification via Dynamic Class Selection (AAAI 2018 · similarity 0.87)
- An Empirical Evaluation of Current Convolutional Architectures' Ability to Manage Nuisance Location and Scale Variability (CVPR 2016 · similarity 0.87)
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.