Mobilenet converted from tensorflow. Training not supported yet.
Download tensorflow checkpoint point from https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet_v1.md and https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/README.md
Then run:
python convert_v1.py --depth_multiplier 0.50 --image_size 224 --tensorflow_model mobilenet_v1_0.50_224.ckpt
This will generate a pth
file which has the same name as the tensorflow checkpoint.
Similar to v2.
(There is currently a mismatch between official mobilenetv2 model code and the official pretrained weights. The last layer in the conv_defs didn't do depth mutliplyication when multiplier is less than 1 according to the weights.)
The preoprocessing follows the tensorflow incecption: transforms.Normalize([0.5, 0.5, 0.5],[0.5, 0.5, 0.5])
.
To match the behavior of tensorflow slim conv2d, we manually calculate the paddings, this may lead to slower speed.
You can download the converted models from link.
Missing.