Issues about the dataset
blueardour opened this issue · comments
Peng Chen commented
Hi, austingg:
Thanks for your sharing of the mobilentv2 network. I want to repeat the experiment and hoping to restire the paper accuracy. There are three questions for me about the dataset.
- Did you employ only the imagenet2012 dataset or also took the imagenet 2016 data when training?
- When resizing the origin images, did you resize them to fixed 256x256 or 256xN. I hear that the later one might result in better accuracy.
- How the data augmentation helps for the accrcury? Is there any comparison?
Yubin Wang commented
@blueardour I haven't use caffe to reproduce the mobilenetv2 result.
2. resize to 256xN, and crop 224x224 for network input
3. inception augmentation helps.
You may reference MXNet's implementation log and scripts.