• 🚨 Michigan University: Deep Learning for Computer vision semester
• Information on bounding boxes: Image Processing Techniques: What Are Bounding Boxes?
  • How Bounding Box Enables Object Detection?
• Information on the metric IoU: Intersection over Union (IoU) for object detection
• AWS Sagemaker:
  • AWS SageMaker in 10 Minutes!
  • AWS Online Tech Talks - Build, Train and Deploy Machine Learning Models on AWS with Amazon SageMaker
  • Step by step to Machine Learning with Sagemaker

• Ane
  • [11-15] July
  • Around 10 days sometime in the beginning of August
• Eldar
• Mohammed
• Sangmeng
  • [11-15] July
• Sebastian
  • [28-31] July

Initial architecture proposal by Ane:

2 main pipelines:

1. Training
   1. Input artifact: testing images
   2. Image preprocessing
   3. Data augmentation
   4. Hyperparameter tuning
   5. Design model
   6. Train model (steps 3 and 4 are repeated multiple times until finding appropriate hyperparameters)
   7. Deploy model
   8. Output artifact: model endpoint
2. Testing
   1. Input artifact: testing images
   2. Image preprocessing (same as in training pipeline)
   3. Call model endpoint
   4. Retrieve and parse results
   5. Output artifacts: test results

Based on test results, the model might be retrained, the preprocessing algorithm modified, etc.

• Yellow boxes are meant to be development steps (data science related)
• Red boxes are meant to be operations steps (AWS related)
  • At least model training and
• Rhombus is for artifacts
  • Blue is input artifacts, images
  • Green is the model endpoint
  • Purple is test results, which we will use to improve our system