Downloads · 30 days
0
MolSight-patent-finetuned
hanmozhang1984/MolSight-patent-finetuned
MolSight-patent-finetuned is a machine learning model from hanmozhang1984. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Fine-tuned MolSight checkpoint for patent chemical structure recognition (OCSR).
Access
Public
Updated Mar 22, 2026
Repo size
778 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pth778 MB · 100%
At a glance
- Access
- Public
- Created
- Mar 22, 2026
- Updated
- Mar 22, 2026
- SHA
- 327ee708
Model card
From the Hugging Face model README
MolSight Patent Fine-tuned
Fine-tuned MolSight checkpoint for patent chemical structure recognition (OCSR).
Performance
Tested on 92 patent structure images from 3 patents:
| Metric | Baseline (PubChem/USPTO) | Fine-tuned |
|---|---|---|
| Exact match | 42.4% | 66.3% |
| Tanimoto >= 0.9 | 71.7% | 91.3% |
| Tanimoto < 0.5 | 4.3% | 0.0% |
Training
- Base model: hustvl/MolSight (pubchem_uspto_smiles_edges_30.pth)
- Fine-tuned on: 1,911 patent SMILES (Indigo-rendered) + 163 real patent images
- Training: SFT, 5 epochs, A10G GPU via Modal
- Data sources: WO2026024861 (bridged bicyclics), WO2020132648 (sulfonamides), WO2025235957 (quinazolines), WO2026011032 (macrocycles)
Usage
The checkpoint auto-downloads from this repo if not found locally.
License
Same as the base MolSight model.
- Created
- Mar 22, 2026
- Updated
- Mar 22, 2026