The original DECIMER repositories had several issues:
- Dependency hell - Installation was problematic due to complex dependencies across different systems
- No self-deployable server - No server version was available
- Slow startup - Loading models on each script invocation was time-consuming
This version solves these by:
- Providing a Docker option for robust, environment-independent deployment
- Server-based architecture with one-time model loading
- Integrated image classifier to filter out non-molecule images early
Warning: Although the classifier can be overridden, it is not recommended. If you have non-molecule images, the system might crash due to the decimer image-to-SMILES implementation.
Note: First-time startup might take a few minutes due to model download (depends on connection speed; happens only once).
Note: Initial recognition calls take time, but batch submissions are much faster after the first call.
GPU note: On Linux/Windows with NVIDIA hardware, GPU acceleration depends on your system/runtime setup (drivers/CUDA/container GPU runtime). This project does not configure NVIDIA stack automatically and will run on CPU when GPU runtime is unavailable. (unless you run the CUDA-enabled Docker build, which requires a compatible NVIDIA setup on the host).
To uninstall, remove the application folder and Python environment, then delete the .data folder in your home directory where Decimer stores its models.