xtanitfy / tf_ssd

1. convert vggnet:

1.1 $cd src/utils && mkdir caffe_model

1.2 put the VGG_ILSVRC_16_layers_fc_reduced.caffemodel and VGG_ILSVRC_16_layers_fc_reduced_deploy.prototxt into caffemodel then run:
  $./ssdVgg2pkl.sh
  [vggnet download] (https://download.csdn.net/download/fireworkpark/10104507)

2.convert dataset:

2.1 $cd data/VKITTI 

2.2 Specify the path of the voc in voc2kitti.py then run:
  $pyhton voc2kitti.py

3.start train:

3.1 ./scripts/train_ssd.sh

4.start eval:

4.1 cd data/VKITTI && python create_test.py 
4.2 ./scripts/eval_ssd.sh
4.3 ./scripts/res_eval.sh (need spicify the train_model)

5.will do:

5.1 Improve accuracy

6.Reference:

6.1 ![caffe-ssd](https://github.com/weiliu89/caffe/tree/ssd)
6.2 ![squeezeDet](https://github.com/BichenWuUCB/squeezeDet)
    .

tf_ssd's Issues

question regarding configuration

Hi
Thanks for sharing your code.
I have a question and appreciate it if you can help me out with it.
in the configaration what is the concepr/meaning of the following variables?

steps = [8, 16, 32, 64, 100, 300]
prior_variance = [0.1, 0.1, 0.2, 0.2]
DECAY_STEPS = 20000
aspect_ratios = [[2], [2, 3], [2, 3], [2, 3], [2], [2]]

I dont understand what is steps and prior variance and decay_steps.
Also im confused about the aspect ratio, why do we define it like [2,3]? shouldn't it be like 1:2 etc?

Thanks