Research map / Detection and segmentation / Open-set and few-shot detection
Disentangling neural mechanisms for perceptual grouping
Junkyung Kim*, Drew Linsley*, Kalpit Thakkar, Thomas Serre
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 horizontal feedback (ablation) | recurrent CNN · Pathfinder | accuracy across difficulty levels | Improved |
| add top-down feedback (ablation) | recurrent CNN · cABC | accuracy across difficulty levels | Improved |
| add both horizontal and top-down feedback (ablation) | recurrent CNN · Pathfinder and cABC | accuracy across difficulty levels | Improved |
| remove horizontal feedback (ablation) | TD+H-CNN · Pathfinder | accuracy as difficulty increases | Hurt |
| remove top-down feedback (ablation) | TD+H-CNN · cABC | accuracy | Hurt |
| remove both horizontal and top-down recurrence (ablation) | TD+H-CNN · Pathfinder and cABC | accuracy | Hurt |
| use one processing timestep instead of recurrent processing (ablation) | H-CNN and TD-CNN · Pathfinder for H-CNN; cABC for TD-CNN | accuracy | Hurt |
| use 8 timesteps instead of 1 timestep (ablation) | fGRU recurrent models · Pathfinder and cABC | performance | Improved |
| replace fGRU modules with conventional GRUs (ablation) | recurrent hierarchical architectures · Pathfinder and cABC | performance | Hurt |
| replace fGRU modules with conventional LSTMs (ablation) | recurrent hierarchical architectures · Pathfinder and cABC | performance | Hurt |
| increase H+TD-CNN depth from 3 to 5 convolutional blocks (ablation) | TD+H-CNN · Pathfinder and cABC | performance | No clear effect |
| increase ResNet-18 width to match ResNet-50 parameter count (ablation) | ResNet-18 · Pathfinder hard | performance | Improved |
Closest papers by meaning
- Vision Transformers Learn Gestalt-Like Figure-Ground Cues from Natural Images (NeurIPS 2026 · similarity 0.89)
- Latent Noise Segmentation: How Neural Noise Leads to the Emergence of Segmentation and Grouping (ICML 2024 · similarity 0.87)
- Flexible Context-Driven Sensory Processing in Dynamical Vision Models (NeurIPS 2024 · similarity 0.87)
- Perceptual Group Tokenizer: Building Perception with Iterative Grouping (ICLR 2024 · similarity 0.85)
- Sparse components distinguish visual pathways & their alignment to neural networks (ICLR 2025 · similarity 0.85)
- Vision CNNs trained to estimate spatial latents learned similar ventral-stream-aligned representations (ICLR 2025 · similarity 0.85)
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.
