Best Molecule Detection Model !
Thank you to open-source. It is the best molecular structure detection model I have ever used!
Much better than MolMiner and other open-sourced molecular structure detection model:
- It leaves enough blank space to avoid cutting off the molecule structure.
- Robust in scanned PDF page images and papers with super complex style.
- Performs well in handwritten molecular detection and small-molecule detection.
I use MolDet YOLO optimized for PDF: "moldet_yolo11l_960_doc.pt"
It is super fast and accurate!
Crazy Perfect !!!
thx!
If you encounter any badcases, you can also provide them here.
Hi, will the training and inference code be available on github? I was curious if adding another class of "coreferences" will be feasible for this model? (I guess some extra data will be required to finetune it)
Hi, will the training and inference code be available on github? I was curious if adding another class of "coreferences" will be feasible for this model? (I guess some extra data will be required to finetune it)
We provide several ultralytics YOLO11 weights. For training codes, you can refer to ultralytics document for details.
We have provided usage in readme: (for inference)
from ultralytics import YOLO
model = YOLO("moldet_yolo11l_640_general.pt")
model.predict("path/to/image.png", save=True, imgsz=640, conf=0.5)
from ultralytics import YOLO
model = YOLO("moldet_yolo11l_960_doc.pt")
model.predict("path/to/pdf_page_image.png", save=True, imgsz=960, conf=0.5)
Adding another class is also possible when fine-tuning our MolDet weights using ultralytics lib.