(Pre)Train AtomLenz

This is the pretraining step of AtomLenz on synthetically generated data with localization labels. The pretrained models (following these steps) are made available here:

• Atom model: https://huggingface.co/spaces/moldenhof/atomlenz/resolve/main/models/atoms_model/synthetic.ckpt
• Bond model: https://huggingface.co/spaces/moldenhof/atomlenz/resolve/main/models/bonds_model/synthetic.ckpt
• Charge model: https://huggingface.co/spaces/moldenhof/atomlenz/resolve/main/models/charges_model/synthetic.ckpt
• Stereo center model: https://huggingface.co/spaces/moldenhof/atomlenz/resolve/main/models/stereos_model/synthetic.ckpt

Download synthetically generated datasets

In order to pretrain the different (atom-level) object detection backbones of AtomLenz, datasets were synthetically generated and are made available here:

• part 1 atom and bond entity annotated images: https://zenodo.org/records/10185264
• part 2 charge and stereocenter entity annotated images: https://zenodo.org/records/10200185

Install ProbKT

As (part of) the training of AtomLenz is based on ProbKT, we refer you to the installation instructions of ProbKT here: https://github.com/molden/ProbKT .

Configuration

Once ProbKT is installed, the downloaded datasets need to be copied (and extracted) to the generate_data folder of ProbKT.

Run first training round (pretraining)

For the pretraining of the RCNN object detection models of AtomLenz the following command needs to be executed for each dataset (atoms/bonds/charges/stereo-centers):

python robust_detection/train/train_rcnn.py --data_path path_to_datasetfolder_inside_generated_data

Run second training round

The second traing round is descriped here: train atomlenz