Research map / Detection and segmentation / Open-set and few-shot detection
Do Deep Neural Networks Suffer from Crowding?
Anna Volokitin, Gemma Roig, Tomaso A Poggio
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 flankers to isolated-target test images | DCNNs and eccentricity-dependent models trained on isolated targets · MNIST target recognition with flankers | accuracy | Hurt |
| Increase the number of flankers | DCNNs and eccentricity-dependent models trained on isolated targets · MNIST target recognition with flankers | accuracy | Hurt |
| Increase target–flanker spacing | DCNNs trained on isolated targets · MNIST targets at fixed eccentricity | accuracy | Improved |
| Train with flankers in the same configuration used at test time | DCNN and eccentricity-dependent models · MNIST target recognition with flankers | accuracy | Improved |
| Train with clutter, then test with different clutter configurations | DCNN and eccentricity-dependent models · MNIST target recognition with unseen flanker configurations | accuracy | Hurt |
| Use more target-similar flankers | DCNNs trained on isolated targets · MNIST targets with MNIST, Omniglot, or notMNIST flankers | accuracy | Hurt |
| Use notMNIST rather than odd MNIST or Omniglot flankers | DCNNs trained on isolated targets · MNIST target recognition with different flanker datasets | accuracy | Hurt |
| Use Omniglot rather than odd MNIST flankers | DCNNs trained on isolated targets · MNIST target recognition with different flanker datasets | accuracy | Improved |
| Add spatial pooling rather than using no total pooling (ablation) | DCNNs · MNIST targets without flankers | accuracy | Improved |
| Add contrast normalization across scales (ablation) | Eccentricity-dependent model · MNIST target recognition across eccentricities | accuracy | Improved |
| Pool across scales at the beginning rather than at the end (ablation) | Eccentricity-dependent model · MNIST target recognition with flankers | accuracy | Hurt |
| Use the eccentricity-dependent model rather than a DCNN | Models trained on isolated targets · MNIST targets embedded in Places images | accuracy | Mixed |
Closest papers by meaning
- Harmonizing the object recognition strategies of deep neural networks with humans (NeurIPS 2022 · similarity 0.86)
- Switching Convolutional Neural Network for Crowd Counting (CVPR 2017 · similarity 0.85)
- Adversarial Margin Maximization Networks (TPAMI 2019 · similarity 0.85)
- Performance-optimized deep neural networks are evolving into worse models of inferotemporal visual cortex (NeurIPS 2023 · similarity 0.85)
- Minimal Images in Deep Neural Networks: Fragile Object Recognition in Natural Images (ICLR 2019 · similarity 0.85)
- Vision Transformers Learn Gestalt-Like Figure-Ground Cues from Natural Images (NeurIPS 2026 · 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.
