training base network (ssd mobilenet v1) about pytorch-ssd

[qfgaohao]

Hi @seongkyun you can download the pertained VGG16 model here
https://s3.amazonaws.com/amdegroot-models/vgg16_reducedfc.pth . This is the one I am using.

Weights initialisation is an open question in NN. There are many research works around it. Generally, it's a good idea to use a pertained model to initialise the weights or part of the weights and make a reasonable start in the early stage of training.

[qfgaohao]

@seongkyun plz see the trained model with the state of the art MP.
https://storage.googleapis.com/models-hao/vgg16-ssd-mp-0_7726.pth

Here is script to train it, just in case you want to train it on your own.
python train_ssd.py --datasets ~/data/VOC0712/VOC2007/ ~/data/VOC0712/VOC2012/ --validation_dataset ~/data/VOC0712/test/VOC2007/ --net vgg16-ssd --base_net models/vgg16_reducedfc.pth --batch_size 24 --num_epochs 200 —scheduler "multi-step” —-milestones “120,160”

[seongkyun]

@qfgaohao Thank you!