search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.73k stars 7.96k forks source link

Problem with getting low mAP #7581

Open jitesh321 opened 3 years ago

jitesh321 commented 3 years ago

I trained my object detection model with both Yolov4 and Yolov4 tiny but in Yolo v4 I am getting very Low mAP i don't know why but I am using the same dataset in both the models but why it's chart_yolov4_tiny yolov4_tiny.txt

not giving good mAP with my dataset. Even in yolov4-tiny, i am not getting a constant mAP as my loss is decreasing but my mAP is somewhat decreasing or remaining constant. can u tell me if this problem with my training parameter or with my dataset?

stephanecharette commented 3 years ago

To me this looks like a normal graph. I don't know why you would call it "low".

jitesh321 commented 3 years ago

but the detection is not working that way I don't know why but even after 78% mAP it's giving a False Positive most of the time

stephanecharette commented 3 years ago

What do your training images look like, and what does your inference image look like?

jitesh321 commented 3 years ago

I am a newbie to this computer vision and Deep learning so I think that this might be low that's why detection is not working in the correct way

stephanecharette commented 3 years ago

No, that graph looks good. Your problem is related to something else. If you are new to this, my guess is your training images and too dissimilar.

stephanecharette commented 3 years ago

Join the discord if you want more help: https://discord.gg/zSq8rtW