Research map / Vision-language and multimodal / Multimodal LLMs
ATOMIC: A Domain-Specific Vision-Language Model for Transmission Electron Microscopy
Chong-ren Tu, HUNG-WEI HSUEH, Shu-han Hsu
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 DomainContext (caption-conditioned) data instead of VisionGround only (ablation) | ATOMIC-7B · TEM-VQA Reasoning | average score: 23.7% -> 25.0% | Improved |
| add DomainContext data instead of VisionGround only (ablation) | ATOMIC-7B · TEM-VQA Methodology | average score: 17.7% -> 18.3% | Improved |
| add DomainContext data instead of VisionGround only (ablation) | ATOMIC-7B · TEM-VQA Perception | average score: 22.2% -> 20.7% | Hurt |
| add DomainContext data instead of VisionGround only (ablation) | ATOMIC-7B · TEM-MCQ Methodology | accuracy: 71.4% -> 73.3% | Improved |
| add DomainContext data instead of VisionGround only (ablation) | ATOMIC-7B · TEM-MCQ Perception | accuracy: 64.8% -> 62.9% | Hurt |
| add DomainContext data instead of VisionGround only (ablation) | ATOMIC-7B · TEM-MCQ Reasoning | accuracy: 86.1% -> 83.0% | Hurt |
| blend VisionGround and DomainContext data versus DomainContext only (ablation) | ATOMIC-7B · TEM-MCQ Perception | accuracy: 62.9% -> 66.5% | Improved |
| scale language model from 7B to 13B (ablation) | ATOMIC-Blend · TEM-MCQ | accuracy: 75.2% to 76.5% | Improved |
| scale language model from 7B to 13B (ablation) | ATOMIC-Blend · TEM-VQA | AWC: 37.5% -> 36.6% | Hurt |
| add downstream task fine-tuning (+FT) (ablation) | ATOMIC-7B-Blend · TEM-MCQ Methodology | accuracy: 73.3% -> 94.5% | Improved |
| add downstream task fine-tuning (+FT) (ablation) | ATOMIC-7B-Blend · MatCha TEM subset (ALL, n=371) | accuracy: 47.7% -> 56.6% | Improved |
| use TEM adaptation (Stage 1/2) initialization before fine-tuning instead of LLaVA-v1.5-7B | ATOMIC-7B-Blend+FT · MatCha TEM subset (ALL, n=371) | accuracy: 38.5 -> 56.6; +18.1 percentage points | 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.
- BLIP-2 — BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation (ICML 2022) · better on 3 of 3 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
- Micro-Bench: A Microscopy Benchmark for Vision-Language Understanding (NeurIPS 2024 · similarity 0.85)
- Fine-Tuned Language Models Generate Stable Inorganic Materials as Text (ICLR 2024 · similarity 0.84)
- LLM Meets Diffusion: A Hybrid Framework for Crystal Material Generation (NeurIPS 2025 · similarity 0.84)
- Learning Vision and Language Concepts for Controllable Image Generation (ICML 2025 · similarity 0.84)
- Periodic Materials Generation using Text-Guided Joint Diffusion Model (ICLR 2025 · similarity 0.84)
- Scalable Diffusion for Materials Generation (ICLR 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.
