Distributed Object detection training - notnamegg/tensorflow-model GitHub Wiki

Usage

Prepare Container

# docker pull notname/tensorflow-model:[tag]

# docker run --runtime=nvidia -it -p 3000:3000 -p 3001:3001 -p 3002:3002
-v [your log dir]:/tmp/trainlog --name [your container name] -d
-v [your dataset dir]:/data/dataset notname/tensorflow-model:[tag]

Modify your config in /data/notnamegg-tensorflow-Distributed-training/models/research/object_detection/samples/configs/[your config]


Run in each host

#docker exec -it [your container name] bash

#cd /data/notnamegg-tensorflow-Distributed-training/models/research

#python object_detection/distributed_train.py --type=[master/ps/worker] --id=0 --master=192.168.1.28 --ps=192.168.1.28 --train_dir=/tmp/train_log --pipeline_config_path=object_detection/samples/configs/[your config]


Result

When all rule host get connected, training will start.

The training result will be in [your log dir] in your host.