How to Evaluate a Yolo Trainning

Hi,i’m a novice on yolo and mxNet.
in my learning has some difficulties,and thanks for helps.

i had got the net “yolo3_darknet53_custom” from “gluoncv.model_zoo”,and trained it on my data set.

the data set i made is similar to handwriting alphabet but just A-F.
the images’ background could be a printed rectangular area,and may include some letters.
i made it 2k per letter,totally 10k pcs image,and label them.

it would be supposed to recognize more than one object.
so i choose object detection net,using the yolo net.

in my training,i get the final obj_loss=0.2.that’s try my best on adjusting parameters.
i wanna for more information to get a better result,such as to get the AP.

so,please,How can i get IOU,MAP,AP in my training script?
best wish.

Hi,

The tutorial for training yolo3 with a gluoncv model can be found here:

https://gluon-cv.mxnet.io/build/examples_detection/train_yolo_v3.html

You can follow this example to create a metric class to compute mAP for your detection dataset.

https://gluon-cv.mxnet.io/_modules/gluoncv/utils/metrics/voc_detection.html

Yeah,
I do that after reading file train_yolo_v3.py etc.
the example voc_detection should be very helpful.
Thanks!