Research map / Detection and segmentation / Open-set and few-shot detection
Recognition in Terra Incognita
Sara Beery, Grant Van Horn, Pietro Perona
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 |
|---|---|---|---|
| test on unseen trans-locations instead of seen cis-locations | Inception-v3 image classifier · CCT-20 | top-1 error: 19.06 -> 41.04 | Hurt |
| test on unseen trans-locations instead of seen cis-locations | Inception-v3 bbox classifier · CCT-20 | top-1 error: 8.14 -> 19.56 | Hurt |
| remove empty images and train on 15 animal classes only (ablation) | Inception-v3 image classifier · CCT-20 trans-locations | top-1 error: 41.08 -> 41.04 | No clear effect |
| add most-confident sequence information (ablation) | Inception-v3 image classifier · CCT-20 trans-locations | top-1 error: 41.04 -> 34.53 | Improved |
| add oracle sequence information (ablation) | Inception-v3 image classifier · CCT-20 trans-locations | top-1 error: 41.04 -> 28.69 | Improved |
| classify cropped bounding boxes instead of full images (ablation) | Inception-v3 · CCT-20 trans-locations | top-1 error: 41.04 -> 19.56 | Improved |
| classify cropped bounding boxes instead of full images (ablation) | Inception-v3 · CCT-20 | cis-to-trans error increase: 115% -> 140% | Hurt |
| add oracle sequence information (ablation) | Inception-v3 bbox classifier · CCT-20 | cis-to-trans error increase: 140% -> 118% | Improved |
| require 10,000 instead of 500 training examples for trans-locations | Inception-v3 classifier · CCT-20 | error rate at 5.33% bbox: 500 -> 10,000 examples | Hurt |
| use Inception-ResNet-v2 instead of ResNet-101 backbone (ablation) | Faster-RCNN · CCT-20 trans-locations | mAP: 70.17 -> 71.37 | No clear effect |
| add most-confident sequence information (ablation) | Faster-RCNN ResNet-101 · CCT-20 trans-locations | mAP: 70.17 -> 84.09 | Improved |
| add most-confident sequence information (ablation) | Faster-RCNN ResNet-101 · CCT-20 | cis-to-trans error increase: 30% -> 4.5% | Improved |
Closest papers by meaning
- Lessons and Open Questions from a Unified Study of Camera-Trap Species Recognition Over Time (ECCV 2026 · similarity 0.89)
- Learning to Detect Novel Species with SAM in the Wild (IJCV 2024 · similarity 0.87)
- WildIng: A Wildlife Image Invariant Representation Model for Geographical Domain Shift (IJCV 2026 · similarity 0.86)
- The SA-FARI Dataset: Segment Anything in Footage of Animals for Recognition and Identification (CVPR 2026 · similarity 0.86)
- MammAlps: A Multi-view Video Behavior Monitoring Dataset of Wild Mammals in the Swiss Alps (CVPR 2025 · similarity 0.86)
- The PanAf-FGBG Dataset: Understanding the Impact of Backgrounds in Wildlife Behaviour Recognition (CVPR 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.
