How to calculate recall? not only AP or mAP

[MuhammadAsadJaved]

Hi, I am using ./darknet map for model evaluation but it only return per class precision , TP and FP values, However I also want to know about recall per class. Even i can not calculate it manually because recall = TP / (TP + FN) and this program only return TP and FP. It also print recall collectively but i want to get recall per class. How can I print recall ? Or at least how Can I get FN values? so that I can calculate it manually.

[amalchaaari]

hello witch command did you use to get the mAP

[MuhammadAsadJaved]

@amalchaaari
I am using different repo (https://github.com/AlexeyAB/darknet) but you can try this command.

$ ./darknet detector map data/obj.data yolo-obj.cfg backup\yolo-obj_7000.weights

or for this repo maybe

$ ./darknet detector test data/obj.data yolo-obj.cfg backup\yolo-obj_7000.weights

Replace .data , .cfg and .weights with yours.

[Adithia99]

i use your command but ,machine said need data/test can you show me data/test ? (the format of data/test)

[MuhammadAsadJaved]

@Adithia99 First check how to download and prepare dataset for training and testing. There are a lot of tutorials on Google. you can use public datasets like coco or voc, there are also tutorials how to prepare your own datasets for yolov3 training for custom object detection.

data/obj.data co contains the path of your dataset , change it where you train.txt and val.txt files located.

it looks somthing like this.

classes= 80 train = ./data/voc/train.txt valid = ./data/voc/val.txt names = data/coco.names backup = backup eval=coco

[Adithia99]

Val.Txt or Data. Test Can you show inside that?

About obj.data mine same with your format.. But still wrong.. I need to see about data/test.txt format (how to make it and like what)

[MuhammadAsadJaved]

it's useless to show mine. it contains thousands images and labels file path for example

/home/computer/data/123.jpg /home/computer/data/123.txt
/home/computer/data/124.jpg /home/computer/data/124.txt
/home/computer/data/125.jpg /home/computer/data/125.txt

these are generated with python shell when you have jpg images and relevent .xml files. , then you will ask how to generate these files. that's why i said first learn from google how to train and test yolov3 on coco or voc dataset.

https://github.com/AlexeyAB/darknet

https://github.com/ultralytics/yolov3

These two darknet repos are well managed , you can also learn from here.

[Adithia99]

Thanks i can do it

[Ashwith-ram]

Hi @MuhammadAsadJaved Did you find out how to calculate precision and recall for individual classes or at least get FN and FP values?