Research map / Vision-language and multimodal / Multimodal LLMs
Insect-Foundation: A Foundation Model and Large Multimodal Dataset for Vision-Language Insect Understanding
Thanh-Dat Truong, Hoang-Quan Nguyen, Xuan-Bac Nguyen, Ashley P. G. Dowling, Xin Li, Khoa Luu
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 attention pooling instead of class token (ablation) | Insect-Foundation ViT-base · IP102 Classification | Acc@1: 72.4 -> 73.3 | Improved |
| add image-text contrastive loss (ablation) | Insect-Foundation ViT-base · IP102 Classification | Acc@1: 73.3 -> 74.2 | Improved |
| add description loss (ablation) | Insect-Foundation ViT-base · IP102 Classification | Acc@1: 74.2 -> 75.8 | Improved |
| add description loss (ablation) | Insect-Foundation ViT-small · IP102 Classification | Acc@5: 89.9 -> 87.7 | Hurt |
| use ViT-base instead of ViT-small backbone (ablation) | Insect-Foundation · IP102 Classification | Acc@1: +4.3% | Improved |
| use ViT-large instead of ViT-base backbone (ablation) | Insect-Foundation · IP102 Classification | Acc@1: +1.1% | Improved |
| use 50% patch sampling instead of 25% (ablation) | Insect-Foundation · IP102 Classification | Acc@1: 65.2 -> 73.3 | Improved |
| use 50% patch sampling instead of 90% (ablation) | Insect-Foundation · IP102 Classification | Acc@1: 69.8 -> 73.3 | Improved |
| train LLaVA on insect instruction data instead of LLaVA v1.5 data (ablation) | LLaVA Vicuna 7B · Insect VQA (IP102-based) | accuracy: 38.14 -> 42.09 | Improved |
| replace CLIP encoder with Insect-Foundation encoder (ablation) | Insect-LLaVA Vicuna 7B · Insect VQA (IP102-based) | accuracy: 42.09 -> 46.08 | Improved |
| use multi-turn instruction data instead of single-turn (ablation) | Insect-LLaVA Vicuna 13B · Insect VQA (IP102-based) | accuracy: 45.57 -> 48.53 | Improved |
| pretrain with Insect-Foundation instead of MAE | ViT-B/16 · IP102 Classification | Acc@1: 72.0 -> 73.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CoCa — COCA: COllaborative CAusal Regularization for Audio-Visual Question Answering (AAAI 2023) · 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
- Insect-Foundation: A Foundation Model and Large-scale 1M Dataset for Visual Insect Understanding (CVPR 2024 · similarity 0.92)
- AgMMU: A Comprehensive Agricultural Multimodal Understanding Benchmark (NeurIPS 2025 · similarity 0.88)
- VLM4Bio: A Benchmark Dataset to Evaluate Pretrained Vision-Language Models for Trait Discovery from Biological Images (NeurIPS 2024 · similarity 0.86)
- AgroOmni: A Large-Scale Multi-view Agricultural Dataset for Cross-Scale Multimodal Reasoning (NeurIPS 2026 · similarity 0.86)
- HoneyBee: Data Recipes for Vision-Language Reasoners (CVPR 2026 · similarity 0.85)
- BIOSCAN-5M: A Multimodal Dataset for Insect Biodiversity (NeurIPS 2024 · similarity 0.84)
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.
