search

ismail-mebsout / Parsing-PDFs-using-YOLOV3

Parsing pdf tables using YOLOV3
112 stars 41 forks source link

darknet command line can't detect correctly? #10

Open hitfishking opened 2 years ago

hitfishking commented 2 years ago

I compiled darknet in cygwin64 on my win10 system, and copied from this project yolov3-tiny_table.cfg，best_v2.weights，a test pdf page image with 2 tables (test.jpg) to the darknet ./cfg, ./, ./data directories respectivly. Then, I typed the following line in the cmd line:

darkent detect cfg/yolov3-tiny_table.cfg  best_v2.weights  data/test.jpg

But，the output didn't detect any table in test.jpg. On the other side, the tables on the test.jpg can be correctly detected using this project! Dose darknet handles .cfg and .weights differently from the way which are used in this project? Could anyone explain this reason?

hitfishking commented 2 years ago

Another question: Did the following image been producted with darknet? If not, which tool was it?

https://github.com/ismail-mebsout/Parsing-PDFs-using-YOLOV3/raw/master/imgs/examples.jpg

hitfishking commented 2 years ago

I get some idea for this question. That is to turn down the confidence thresh parameter of darknet. By default, Yolo only displays objects detected with a confidence of 0.25 or higher. When I turn down the thresh to 0.1，tables on the test.jpg are detected!

 darkent detect cfg/yolov3-tiny_table.cfg  best_v2.weights  data/test.jpg -thresh 0.1