low mAP on coco val2014

[janetwise]

Has anyone been able to reproduce the mAP of ~55 by running:
python3 test.py --weights_path weights/yolov3.weights
I was only able to get mAP 50? I got mAP 53 by setting the image size to 608.

I just cloned the repo today and downloaded the yolov3 weights as shown in the script.

I got mAP 29 using the yolov3_tiny.weights. Is it in the right range?
Thank in advance!

[ThanhPham1987]

me too

[jinlintansjtu]

me too

[lovedoubledan]

me too

[hank08tw]

me too I only got mAP: 0.5145196513137332 map

[yanxiaod123]

me too, why?

[LeungWaiHo]

me too, I think if it has contingency?

[qqsuhao]

I clone the code on 2020-03-31 and run the "test.py" with yolov3.weights, but I just get mAP: 0.3794739921856155.

[Flova]

What confidence threshold did you use? The yolo paper uses afaik relatively low thresholds such as 0.001 combined with another test dataset called test-dev.

[yjy4231]

What confidence threshold did you use? The yolo paper uses afaik relatively low thresholds such as 0.001 combined with another test dataset called test-dev.

How can I get the test-dev ?

[Flova]

Have a look at this list https://cocodataset.org/#download