Research map / Vision-language and multimodal / Cross-modal learning
BioVITA: Biological Dataset, Model, and Benchmark for Visual-Textual-Acoustic Alignment
Risa Shinoda, Kaede Shiohara, Nakamasa Inoue, Kuniaki Saito, Hiroaki Santo, Fumio Okura
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 Stage 2 tri-modal training after Stage 1 (ablation) | BioVITA · seen species-level retrieval | average Top-1: 64.3 -> 71.7 | Improved |
| add Stage 1 audio-text training (ablation) | BioVITA · seen species-level retrieval | average Top-1: 34.2 -> 71.7 | Improved |
| initialize from pretrained BioCLIP 2 (ablation) | BioVITA Stage 1 · seen species-level retrieval | average Top-1: 25.1 -> 64.3 | Improved |
| make the text encoder trainable (ablation) | BioVITA · seen species-level retrieval | average Top-1: 64.5 -> 71.7 | Improved |
| add image-text contrastive loss (ablation) | BioVITA · seen species-level retrieval | average Top-1: 64.9 -> 71.7 | Improved |
| add audio-image contrastive loss (ablation) | BioVITA · seen species-level retrieval | average Top-1: 64.5 -> 71.7 | Improved |
| use scientific name instead of common name in prompt | BioVITA · seen species-level retrieval | average Top-1: 71.7 -> 73.5 | Improved |
| train on full data instead of 25% (ablation) | BioVITA Stage 1 (Sci prompt) · seen species-level retrieval | average Top-1: 53.8 -> 64.3 | Improved |
| train on full data instead of 50% (ablation) | BioVITA Stage 1 (Sci prompt) · seen species-level retrieval | average Top-1: 57.2 -> 64.3 | Improved |
| move from Stage 1 to Stage 2 | BioVITA · unseen species-level retrieval | average Top-1: 19.8 -> 51.9 | Improved |
| move from species to family level retrieval | BioVITA Stage 2 · seen retrieval | average Top-1: 71.7 -> 36.4 | Hurt |
| Stage 2 training vs BioCLIP 2 on image-text retrieval | BioVITA · seen species-level I2T | Top-1: 65.1 -> 86.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- ImageBind — ImageBind: One Embedding Space To Bind Them All (CVPR 2023) · better on 6 of 6 shared results
- BioCLIP 2 — BioCLIP 2: Emergent Properties from Scaling Hierarchical Contrastive Learning (NeurIPS 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- NatureLM-audio: an Audio-Language Foundation Model for Bioacoustics (ICLR 2025 · similarity 0.90)
- BioCLIP: A Vision Foundation Model for the Tree of Life (CVPR 2024 · similarity 0.88)
- The iNaturalist Sounds Dataset (NeurIPS 2024 · similarity 0.87)
- BioCLIP 2: Emergent Properties from Scaling Hierarchical Contrastive Learning (NeurIPS 2025 · similarity 0.87)
- VLM4Bio: A Benchmark Dataset to Evaluate Pretrained Vision-Language Models for Trait Discovery from Biological Images (NeurIPS 2024 · similarity 0.87)
- BioTrove: A Large Curated Image Dataset Enabling AI for Biodiversity (NeurIPS 2024 · similarity 0.87)
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.
