How to train your own Object Detector with TensorFlow’s Object Detector API
Dat Tran
7.4K193

Hi Dat,

I have followed steps present in the article for my images of 7 classes. I started to train it in local but the process was getting killed due to memory issues. Therefore I decided to move to google cloud. I have followed steps present in https://cloud.google.com/sdk/downloads#versioned , https://cloud.google.com/ml-engine/docs/command-line,

I have created required tar packages in my local file : dist/object_detection-0.1.tar.gz,slim/dist/slim-0.1.tar.gz

I am in trial period as of now. I created new project and two buckets:

+ data/
- model.ckpt.index
- model.ckpt.meta
- model.ckpt.data-00000-of-00001
- pascal_label_map.pbtxt
- pascal_train.record
- pascal_val.record
+ config/
- ssd_mobinet_v1_pets.config

I am executing following command from From tensorflow/models/research/ folder:


gcloud ml-engine jobs submit training object_detection_26_10 --job-dir=gs://data --packages dist\object_detection-0.1.tar.gz,slim/dist/slim-0.1.tar.gz --module-name object_detection.train --config training/cloud.yml -- --train_dir=gs://data --pipeline_config_path=gs://config/ssd_mobilenet_v1_pets.config

I am getting error: code 403:

Google search suggest , i have some billing related issues, but i have enable billing for this project.

Have you ever faced similar issue with GCP.

Many thanks in advance.