Joshua Z. Zhang

Seems like I don't have one right now. Once I get my hands on them, I will update.

awesome, nice work.

use matplotlib savefig function

use ``` git clone --recursive https://github.com/zhreshold/mxnet-yolo cd mxnet-yolo/mxnet # try to build it using the official installation tutorial ``` don't trick submodule from offical repo

@jianglong98 You have to build from source, as I have mentioned the official repo is lacking some operators, pip, docker won't work at all.

This is the padding for label because labels can have arbitrary size. You can increase the pad-width manually.

specify label_width manually: https://mxnet.incubator.apache.org/api/python/io/io.html?highlight=imagedetrecorditer#mxnet.io.ImageDetRecordIter

Did you use pretrained models? Inception should be faster than VGG regardless of CPU/GPU devices.

http://newdocs.readthedocs.io/en/latest/distributed_training.html

Add a kv_store to Module, and split iterator using num_parts.