aws-samples / amazon-sagemaker-aws-greengrass-custom-object-detection-model Goto Github PK

End to end code samples for training an object detection model in Amazon SageMaker using the built-in SSD algorithm and running it on AWS IoT Greengrass.

License: MIT No Attribution

Python 1.25% Jupyter Notebook 98.75%

greengrass lambda s3 sagemaker

amazon-sagemaker-aws-greengrass-custom-object-detection-model's Introduction

Training the Built-In Object Detection Model in Amazon SageMaker and running it on AWS IoT Greengrass

End to end code samples for training an object detection model in Amazon SageMaker using the built-in SSD algorithm and running it on AWS IoT Greengrass.

Object detection is the process of identifying and localizing objects in an image. A typical object detection solution takes in an image as input and provides a bounding box on the image where an object of interest is, along with identifying what object the box encapsulates.

Many scenarios of object detection happen in places with limited connectivity/bandwidth to internet. Therefore, running object detection at the IoT Edge is a often a solution in these use cases.

This repo contains useful scripts and Juypter notebooks from collecting training data from a webcam to data labeling, to building an object detection model using built-in SSD model from Amazon SageMaker, and finally, deploying it to run and make inference the edge using AWS IoT Greengrass.

Accompanying Blog posts

This repo supports the 3-part blog post on AWS IoT blog: Training the Amazon SageMaker object detection model and running it on AWS IoT Greengrass

Architecture

Sections

The repo is organized into 3 sections:

The scripts in each of these folders are prefixed with a number. The number in the prefix represents the order the scripts should be used in. E.g.

data-prep/
├── 00_get_video.py
├── 01_video_to_frame_utils.py
├── 02_generate_gt_manifest.py
├── 03_visualize_gt_labeling_manifest.py
├── 04_create_ground_truth_job.ipynb
├── 05_visualize_ground_truth_labels.py

License Summary

This sample code is made available under the MIT-0 license. See the LICENSE file.

amazon-sagemaker-aws-greengrass-custom-object-detection-model's People

Contributors

Stargazers

Watchers

amazon-sagemaker-aws-greengrass-custom-object-detection-model's Issues

Appending Additional Metadata: Unable to append in my own model solution

In appending additional metadata, you guys only gave an example for your demo. You listed it as optional for anyone who is making their own model, but it is a necessary step. You should provide instructions on how to append it for your own.

I had trouble because instead of two additional metadata's I had one.

I solved it by going into your script, data-prep/02_generate_gt_manifest.py, and changed the additional metadata's from object and color to one "class".

You should give specific instructions or in least a pointer for other people following the guide.

GG V2

Is there a greengrassV2 version available for this sample code?

"Convert the trained model artifact to a deployable model artifact" error when try built-in object detection model

Hi, I'm following Training the Amazon SageMaker object detection model and running it on AWS IoT Greengrass – Part 2 of 3: Training a custom object detection model to see if my trained model artifacts can work with my IOT device via greengrass.

In the "Convert the trained model artifact to a deployable model artifact" section by walking through the ipynb, line 2 !pip install gluoncv show error ERROR: Could not find a version that satisfies the requirement autocfg (from gluoncv) (from versions: none) ERROR: No matching distribution found for autocfg (from gluoncv). Do you have any suggest on this?

I'm running the ipynb in sagemaker notebook ml.t2.medium with conda_amazonei_mxnet_p27 kernel.

Deploy step failing for SSD object detection model while getting ready to inference model locally.

Hi @angelarw, @zhreshold

I was following the below notebook to inference sagemaker object detection ssd model locally. I get an error reporting that commons module does not exist when executing the deploy step to make the model ready for deployment.

https://github.com/aws-samples/amazon-sagemaker-aws-greengrass-custom-object-detection-model/blob/master/training/03_local_inference_post_training.ipynb

Error:

Join together outputs from multiple labeling jobs

Hi, I really liked this sample and it helped me an awfully lot. I am quite a newbaby to AWS so I would like to have one question regarding section of Join together outputs from multiple labeling. I have multiple labeling jobs and I would like to join labeled data from this jobs into one bigger dataset, but the data are stored in different s3 buckets. Do you think, it is possible, to join even these datasets?
Because when I try it, I always get errors while training, so I am not sure, if it is even possible.
Thank you very much for you response, I would greatly appreciate it.

Getting Syntax error: word unexpected (expecting ")") when running $ sudo ./greengrassd start

Hello,

Everything has been ok for now, following the installation so far. However, when trying to start the greengrass deamon, I am getting a Syntax error: word unexpected (expecting ")") when running $ sudo ./greengrassd start . I have no clue what it is and have tried to find a solution online with no results. Any help would be appreciated. Thank you.