search

MahmudulAlam / Automatic-Identification-and-Counting-of-Blood-Cells

Machine learning approach of automatic identification and counting of blood cells (RBC, WBC, and Platelet) with KNN and IOU based verification.
https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/htl.2018.5098
GNU General Public License v3.0
133 stars 53 forks source link

testing on high-resolution image #5

Closed GloriaZLQ closed 5 years ago

GloriaZLQ commented 5 years ago

when you use the model on high-resolution image and divide image into 3x3grid, how to process the cells at the edge of the image.

MahmudulAlam commented 5 years ago

@GloriaZLQ Hi. Model is used on the high-resolution image in Prediction_on_HRI.py and partially visible cells at the edge are ignored by the model as in the training set edge cells are not annotated as ground truth cells.

MahmudulAlam commented 5 years ago

The Images\ folder was missing high-resolution images previously. I have added them to the folder.

GloriaZLQ commented 5 years ago

I watched your result in README "Combined Output",found that one rbc cell was counted twice and some rbc cells were ignored at the edge of the 3x3 grid. I think this problem affects the counting accuracy and do you have any solution?

MahmudulAlam commented 5 years ago

That RBC cell is divided by grid line and two predictions are coming from two grid cells. Double counting can be prevented using proposed KNN and IOU based verification which is implemented only for platelets only. Those high-resolution images (HRI) are used only to verify the generalization of the learning and not used any similar image during training and HRI dataset only contains images, not any annotations. So, if you want to show any counting performance in those images, you have to manually annotate the cells first. Edge cells are ignored because in the training dataset edge cells are not considered as full cell and not annotated as ground truth cell.

GloriaZLQ commented 5 years ago

ok,thank you