About Documents

• Training on Colab and Instructions for Colab setup
• Training on VM
  • more on Google VM
• Preparing Data for Training & File Formats
• Models for Retraining
• Training Configuraton/Parameters
• Input/Output Sourcing Options
• Github Data hosting
• About Checkpoints and more
• Tensorboard ---must clear/move event files in between training sessions. See official tensorboard github for documentation on how to view multiple runs
• Evaluation (object det with tf2 only)
• Understanding:
  • Dropout: The Dropout layer randomly sets input units to 0 with a frequency of rate at each step during training time, which helps prevent overfitting. Inputs not set to 0 are scaled up by 1/(1 - rate) such that the sum over all inputs is unchanged.
• Explore Deep Articles from Students:
  • Parameter Turning (batchsize, learning rate +)
  • Overtraining + Importance of Checkpoints
  • EfficientDet D4
  • Attention Mechanism and see Tesnorflow example to train model using attention for Image Captioning (input single image, and process image first with an Inception-based feature extractor)