mAP won't go above 50 and avg loss stays around 1.4/1.5

[roelofd]

I want to train Darknet tiny YOLO to recognize playing cards. Some info: Dataset = 1000 images Classes amt = 7 mAP test set amt = 200

my tiny yolo cfg is as follows: `[net]

Testing

batch=64

subdivisions=8

Training

batch=64 subdivisions=8 width=608 height=608 channels=3 momentum=0.9 decay=0.0005 angle=0 saturation = 1.5 exposure = 1.5 hue=.1 flip=0

learning_rate=0.0001 burn_in=1000 max_batches = 14000 policy=steps steps=11200,12600 scales=.1,.1`

and my filter values above the {yolo] layers is 36 I have not recalculated the anchors and the weight i used is "yolov3-tiny.conv. I have done about 1500 iterations at this point and since 1000 iterations there is barely any chance in results.

Is there anybody that knows what the problem might be? Could this possibly my dataset?

[kingjam3s]

I had a problem with the mAP and the problem was the dataset. but maybe you have a different problem...

[harsh52]

There may be several reasons for that, if you have done everything right and your test dataset size is sufficient enough then i will highly recommend you to look at this tutorial for fine tune your object detection. https://github.com/jaskarannagi19/yolov3/blob/master/README.md#how-to-train-to-detect-your-custom-objects

[hasansalimkanmaz]

I have reached 54 mAP for a different dataset after 3500 iterations and applying different learning rates.