Research map / Vision-language and multimodal / Cross-modal learning
Bioacoustic Geolocation: Species Sounds as Geographic Signals
Mustafa Chasmai, Wuao Liu, Subhransu Maji, Grant Van Horn
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 |
|---|---|---|---|
| pretrain audio encoder on iNatSounds instead of ImageNet (ablation) | AG-CLIP · iNatSounds test | region accuracy: 14.4 -> 17.2 | Improved |
| pretrain audio encoder on iNatSounds instead of ImageNet (ablation) | AG-CLIP · iNatSounds test | continent accuracy: 64.9 -> 71.2 | Improved |
| add auxiliary checklist loss to GeoCLIP location encoder (ablation) | AG-CLIP · iNatSounds test | continent accuracy: 68.5 -> 71.2 | Improved |
| use GeoCLIP location encoder instead of SatCLIP (ablation) | AG-CLIP · iNatSounds test | region accuracy: 14.7 -> 17.0 | Improved |
| use average pooling over clips instead of max pooling (ablation) | AG-CLIP · iNatSounds test | region accuracy: 16.5 -> 17.2 | Improved |
| use average pooling instead of K-means cluster pooling (ablation) | AG-CLIP · iNatSounds test | region accuracy: 15.5 -> 17.2 | Improved |
| use trained transformer pooling instead of averaging (ablation) | AG-CLIP · iNatSounds test | continent accuracy: 71.2 -> 71.5 | No clear effect |
| use hierarchical classification instead of single-resolution Res-2 (ablation) | Classification · iNatSounds test | country accuracy: 36.5 -> 40.0 | Improved |
| weight SINR maps by all predicted species instead of top-1 species (ablation) | Species ranges · iNatSounds test | region accuracy: 08.2 -> 09.3 | Mixed |
| use Haversine instead of Euclidean regression loss (ablation) | Regression · iNatSounds test | region accuracy: 2.9 -> 4.2 | Improved |
| aggregate recordings by Res-5 hexagon over a year | AG-CLIP · iNatSounds test | region accuracy: 17.2 -> 30.4 | Improved |
| aggregate recordings by Res-6 hexagon over a week instead of a year (ablation) | AG-CLIP · iNatSounds test | region accuracy: 25.9 -> 19.9 | Hurt |
Closest papers by meaning
- NatureLM-audio: an Audio-Language Foundation Model for Bioacoustics (ICLR 2025 · similarity 0.84)
- The iNaturalist Sounds Dataset (NeurIPS 2024 · similarity 0.83)
- BAT: Learning to Reason about Spatial Sounds with Large Language Models (ICML 2024 · similarity 0.83)
- RANGE: Retrieval Augmented Neural Fields for Multi-Resolution Geo-Embeddings (CVPR 2025 · similarity 0.82)
- MRSAudio: A Large-Scale Multimodal Recorded Spatial Audio Dataset with Refined Annotations (NeurIPS 2025 · similarity 0.82)
- PIGEON: Predicting Image Geolocations (CVPR 2024 · similarity 0.82)
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.
