Vathos 🐲 - The Journey

Q : So what does the model do ? Ans: It takes in 2D images and outputs segmented image + a depth map, basically recognizes the object for which it was trained for, segments it out, and also creates a 3D depth map of the image.

Read through The Journey of how the dataset was made, how the model was researched and finally how it was trained, if you want to see the results of the model, head over to Training on Small Dataset, a few more model outputs can be found in Loss Functions.

Contents:

• 1. The Model
  • Model Architecture
  • Summary of the Model
• 2. The Dataset
  • Dataset Google Drive link
  • Dataset Creation
  • Depth Map creation
  • Mean and Standard Deviation
  • Dataset Visualization
  • How the dataset was created
• 3. Loss Functions
  • Playing with Loss Functions
• 4. Testing Timings and Hyper Params
  • Playing with batch_size
  • Testing Model Timings
• 5. Run on TPU !
• 6. Training on Small Dataset
  • LR Range Test
  • Train for 15 Epochs
  • Results
• 7. Dataset and Model Exploration Log
  • Ideas
  • TODO
  • Model
  • Requirements
  • Notes
  • Colab Accounts Management
  • Further Improvements
  • Targets