Dual-path Proposal Discriminative Network for abnormal cell detection in cervical liquid-based cytology images.

Framework:

Source Code Project:

--checkpoints: to save output files (checkpoints, tensorboard, log files, etc.) and pretrained models.

--demo: input and output images for "demo.py"

--detectron2: see "https://github.com/facebookresearch/detectron2."

--models: network architectures of DPD-Net, including backbone (att_fpn.py), RoIHead with Proposal Relation Module (PRM) (relation_roi_heads.py), detection networks (single_cell_rcnn.py, dualway_rcnn.py).

--tools: files for training, testing, visualizing.

--utils: calculation of Proposal Contrastive Loss (PCL).

Demo:

input images & output images: