Caffe implementation of SSD detection on MobileNetv2, converted from tensorflow.
- Firstly you should download the original model from tensorflow.
- Use gen_model.py to generate the train.prototxt and deploy.prototxt (or use the default prototxt).
python gen_model.py -s train -l labelmap_coco.prototxt -d trainval_lmdb -c 91 >train.prototxt
python gen_model.py -s deploy -l labelmap_coco.prototxt -d trainval_lmdb -c 91 >deploy.prototxt
- Use dump_tensorflow_weights.py to dump the weights of conv layer and batchnorm layer.
- Use load_caffe_weights.py to load the dumped weights to deploy.caffemodel.
- Use the code in src to accelerate your training if you have a cudnn7, or add "engine: CAFFE" to your depthwise convolution layer to solve the memory issue.
- Maybe you need deploy.caffemodel for VOC dataset, use coco2voc.py to get it.
There are some differences between caffe and tensorflow implementation:
- The padding method 'SAME' in tensorflow sometimes use the [0, 0, 1, 1] paddings, means that top=0, left=0, bottom=1, right=1 padding. In caffe, there is no parameters can be used to do that kind of padding.
- MobileNet on Tensorflow use ReLU6 layer y = min(max(x, 0), 6), but caffe has no ReLU6 layer.
The original/deploy.prototxt shows a exactly same version with tensorflow, and you can use my tensorflow-compatible ssd to see the result. After that, I will do some finetuning on the converted model to make it work on original ssd in the next few days.