Tensorflow-Object-Detection-Api Docker Files

With Github and Docker's help, I solve a really frustating problem, my problem was each time I make some changes to a library or tensorflow version the tensorflow object detection API didnt produce any output. There were n number of errors and it wasted alot of my time to fix them each time. Finally I made a Docker container whos instructions are given in nice details below, and now whenever I have a trained model from object detection API. I just build docker image and test it and it tests models with good consistency. . Note I use object_detection2.ipynb from research/object-detection directory (which is inside the docker container) to test my retrained models.

To build the docker file navigate to tensorflow1.14_cpu_p36 folder and type:

docker build -t tensorflow .

After building start: (without assigning IP):

docker run <container name>

After building start: (with assinment of IP):

docker run --name tensorflow -p 8888:8888 -d tensorflow

And open http://localhost:8888 Password: root

To ssh into a docker container

docker exec -it <container-id> sh

docker exec -it <container name> /bin/bash

another way to login

sudo docker run -i -t <container id> /bin/bash

sudo docker run -i -t <container id>

To stop the docker container:

docker container stop <container name>

To remove the docker container:

docker container rm <container name>

To copy files from localhost to docker

docker cp <local_folder> <docker_container_name>:<full_path of folder inside docker>

note <docker_container_name>: can be got by command : docker ps -a

This command converts the container 9c09acd48a25 to an image with the name fedora-base-with-git:

docker commit 9c09acd48a25 fedora-base-with-git

To copy files from docker to localhost

docker cp <docker_container_name>:<full_path of folder inside docker> <local_folder>

https://ajinkya93.atlassian.net/l/c/qVxC3MeK

ajinkya933 / Tensorflow_Api_Docker_Files