Darknet: Training YOLO3 on 16 classes VOC --> error

[austin880301]

Dear sir:

I modify the voc_lable.py for 16 classes and change yolov3-voc-cfg for classes=16, filters=63. and then train the voc.
At last run detect demo with my new weights and the result as following:

FPS:20.1
Objects:

* Error in `./darknet': free(): invalid pointer: 0x00007f2ff80023d0 *
======= Backtrace: =========
/lib/x86_64-linux-gnu/libc.so.6(+0x777e5)[0x7f30fb7817e5]
/lib/x86_64-linux-gnu/libc.so.6(+0x8037a)[0x7f30fb78a37a]
/lib/x86_64-linux-gnu/libc.so.6(cfree+0x4c)[0x7f30fb78e53c]
./darknet[0x4672e8]
./darknet[0x485597]
/lib/x86_64-linux-gnu/libpthread.so.0(+0x76ba)[0x7f30fbadb6ba]
/lib/x86_64-linux-gnu/libc.so.6(clone+0x6d)[0x7f30fb81141d]
======= Memory map: ========
......

best regards

[ahsan856jalal]

Did you change filters and classes on all three instances in the cfg file

[austin880301]

Thanks a lot , I do not change filters and classes on all three instances in the cfg file. I only change one yolo layer, I should change three yolo layer

[zhangjinsong3]

@austin880301 ，Excuse me. How to set the 'filters' according to 'classes'

[Ibtastic]

@zhangjinsong3
Each YOLO layer has 255 outputs: 85 outputs per anchor [4 box coordinates + 1 object confidence + 80 class confidences], times 3 anchors. If you use fewer classes, reduce filters to filters=[4 + 1 + n] * 3, where n is your class count. This modification should be made to the layer preceding each of the 3 YOLO layers. Also modify classes=80 to classes=n in each YOLO layer, where n is your class count (for single class training, n=1)

As mentioned in this https://github.com/ultralytics/yolov3/issues/102