MolVision Molecular Image Recognition (OCSR)

Uses PatSight MolVision to perform optical chemical structure recognition (OCSR) on molecular structure images, extract SMILES/SDF, compute molecular properties by RDKit, and generate visualization dashboard image.

Authentication

This skill calls the PatSight Patent Extractor API and requires patsight account and password.

First-time users: Register at PatSight.

Add to your OpenClaw config:

{
  "skills": {
    "entries": {
      "ocsr": {
        "enabled": true,
         "env": {
          "PATSIGHT_ACCOUNT": "PATSIGHT_ACCOUNT",
          "PATSIGHT_PASSWORD": "PATSIGHT_PASSWORD,
        },
      }
    }
  }
}

Or set environment variables:

export PATSIGHT_ACCOUNT="your_account"
export PATSIGHT_PASSWORD="your_password"

Quick Start

Using the Script

uv run python skills/ocsr/scripts/run_ocsr.py patent <image1> <image2> ...
uv run python skills/ocsr/scripts/run_ocsr.py mol1.png mol2.png --outdir ./ocsr_output
uv run python skills/ocsr/scripts/run_ocsr.py patent image.jpg --token <token>

Examples

# Basic OCSR on images (backward compatible: omit "patent" subcommand)
uv run python skills/ocsr/scripts/run_ocsr.py mol1.png mol2.png

# With output directory
uv run python skills/ocsr/scripts/run_ocsr.py patent mol1.png mol2.png --outdir ./ocsr_output

# With token
uv run python skills/ocsr/scripts/run_ocsr.py patent image.jpg --token $PATSIGHT_TOKEN

# With account credentials
uv run python skills/ocsr/scripts/run_ocsr.py patent image.png --account user --password pass

Options

Supported Image Formats

.png, .jpg, .jpeg, .webp, .bmp, .tif, .tiff

Output

• ocsr_result_<timestamp>.json: Raw API response + enriched RDKit properties
• ocsr_dashboard_<timestamp>.png: Visualization report

Each result includes smiles, sdf_str, score, and RDKit-derived properties (formula, MW, LogP, TPSA, etc.).

API Reference

Endpoint

POST {base_url}/v1/u/extractor/image/molecules?request_id=<id>

• base_url: Default https://patent.xinsight-ai.com/patent/api
• Requires Authorization: <token> header

Multipart fields

• image: Image file(s) (one image field per file, multiple allowed)
• label: Fixed as molvision

Response format

API returns {"code":1, "data":{"data":[...], "msg":"SUCCESS", ...}}. The script converts it to {"data": [...]} for downstream processing. Each record contains smiles, sdf_str, score, idx, page_idx, rect, etc.

Tips

• 1-N images: Accept multiple image files in one run
• Token priority: Command-line --token > PATSIGHT_TOKEN > --account/--password > env vars
• RDKit: If parsing fails, raw result is preserved and properties marked unavailable
• Partial failure: If some files fail, processing continues and reports partial success