Eric Liu

Refer this link : https://stackoverflow.com/questions/2987207/why-do-i-get-a-c-malloc-assertion-failure

1. [Make lmdb](https://github.com/eric612/MobileNet-YOLO/tree/master/data/cityscapes) 2. Use these [models](https://github.com/eric612/MobileNet-YOLO/tree/master/models/cityscapes) to train However , it is not complete tutorial , you may meet some problems , it is welcome to ask question here.

@kkjong If you want to train models , recommend to use imagenet pre-trained. And the channel numbers was based on your [lablemap setting](https://github.com/eric612/MobileNet-YOLO/blob/master/data/cityscapes/labelmap_seg.prototxt) But here has a problem , the...

https://drive.google.com/drive/folders/1KhgK25ct7C02a31XmVgWcoeFLs_yBXdg deploy 跟 caffemodel 可以搭配 ， train and test 需要搭配前一版的[prototxt](https://github.com/eric612/MobileNet-YOLO/tree/5b4bba5ff36053e76b514f61cf43b2536c2b2f6b/models/mobilenetv2_voc) ，差別在於我把batchnorm 跟scale 改成replace，方便merged ，我目前還在訓練COCO dataset ，模型還沒finalize

In docker build , I use opencv 4.x version and ubuntu18 , and it work fine https://hub.docker.com/r/eric612/mobilenet-yolo

Can you show me the training logs or send it to me ?

The gt box number are unbalance at each scale layer , like below , you may need to generate new anchors like [wiki](https://github.com/eric612/MobileNet-YOLO/wiki) , link "Generate anchors to increase performance"...

It recommend to use 416 , you can try to use 416 first

Use this [parameters](https://github.com/eric612/MobileNet-YOLO/blob/master/models/mobilenetv3_voc/train.prototxt#L3999) with each yolo layer

Try to change the batch size in test prototxt