Research map / Label-efficient and robust learning / Recognition and long-tail
Insect Identification in the Wild: The AMI Dataset
Aditya Jain, Fagner Cunha, Michael J Bunsen, Juan Sebastián Cañas, Léonard Pasi, Nathan Pinoy, Flemming Helsing, JoAnne Russo, Marc S Botham, Michael Sabourin, Jonathan Fréchette, Alexandre Anctil and 16 more
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 mixed-resolution (MixRes) augmentation on top of RandAugment and random resize crop (ablation) | ResNet50 · AMI-Traps NE-America species | micro top-1 accuracy (%): 67.88 -> 71.86 | Improved |
| add MixRes without other augmentation (ablation) | ResNet50 · AMI-Traps NE-America species | micro top-1 accuracy (%): 51.46 -> 58.41 | Improved |
| add MixRes (ablation) | ResNet50 · AMI-GBIF NE-America test | micro accuracy (%): 89.71 -> 89.39 | Hurt |
| add RandAugment and random resize crop (ablation) | ResNet50 · AMI-Traps NE-America species | micro top-1 accuracy (%): 51.46 -> 67.88 | Improved |
| raise input resolution from 128 to 224 (ablation) | ResNet50 · AMI-Traps NE-America species | micro top-1 accuracy (%): 71.86 -> 68.41 (AMI-GBIF 89.39 -> 92.97) | No clear effect |
| pretrain on all AMI-GBIF with ImageNet weights before regional fine-tuning (ablation) | ResNet50 · AMI-Traps NE-America species | micro top-1 accuracy (%): 71.86 -> 75.06 | Improved |
| add square padding of crops at inference (ablation) | ResNet50 · AMI-Traps binary (all crops) | precision (%): 66.21 -> 68.31 | Improved |
| use ConvNeXt-B instead of ResNet50 | fine-grained classifier · AMI-GBIF NE-America few-shot species | micro accuracy (%): 34.03 -> 45.81 | Improved |
| use ViT-B/16 on C-America in-distribution | fine-grained classifier · AMI-GBIF C-America species test | micro accuracy (%): 88.62 -> 79.80 (ConvNeXt-B vs ViT-B/16) | Hurt |
| train region model with about 10 times fewer images (C-America) | fine-grained classifier · AMI-Traps C-America genus | micro top-1 accuracy (%): 85.46 -> 51.92 (NE-America vs C-America ConvNeXt-B) | Hurt |
Closest papers by meaning
- Open-Insect: Benchmarking Open-Set Recognition of Novel Species in Biodiversity Monitoring (NeurIPS 2025 · similarity 0.93)
- Insect-Foundation: A Foundation Model and Large-scale 1M Dataset for Visual Insect Understanding (CVPR 2024 · similarity 0.91)
- BIOSCAN-5M: A Multimodal Dataset for Insect Biodiversity (NeurIPS 2024 · similarity 0.90)
- BioTrove: A Large Curated Image Dataset Enabling AI for Biodiversity (NeurIPS 2024 · similarity 0.88)
- BuckTales: A multi-UAV dataset for multi-object tracking and re-identification of wild antelopes (NeurIPS 2024 · similarity 0.86)
- DivShift: Exploring Domain-Specific Distribution Shifts in Large-Scale, Volunteer-Collected Biodiversity Datasets (AAAI 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.
