- 🚨 Michigan University: Deep Learning for Computer vision semester
- Information on bounding boxes: Image Processing Techniques: What Are Bounding Boxes?
- Information on the metric IoU: Intersection over Union (IoU) for object detection
- AWS 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:
- Training
- Input artifact: testing images
- Image preprocessing
- Data augmentation
- Hyperparameter tuning
- Design model
- Train model (steps 3 and 4 are repeated multiple times until finding appropriate hyperparameters)
- Deploy model
- Output artifact: model endpoint
- Testing
- Input artifact: testing images
- Image preprocessing (same as in training pipeline)
- Call model endpoint
- Retrieve and parse results
- 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