Skip to content

Multi-Headed Variational Autoencoder using Residual Networks, for Self Driving

Notifications You must be signed in to change notification settings

css459/dl-final-project

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

53 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Residual Variational Autoencoder (resvar)

Deep Learning, Center for Data Science, NYU

The inner workings of this model are documented in detail in doc/report.pdf

NEW: Feature Pyramid Networks (branch resnet-fpn)

This resvar implementation has been overhauled using the Dense Transformer Networks and Feature Pyramid Networks proposed by Roddick and Cipolla, 2020. This implementation also comes with a custom pretraining task of predicting the permutation of the six input images.

This model is not documented in the report

model

Contains the resvar model in resnet along with other helper models like segmentation and a plain vae that was also implemented. A util class implements intermediate models.

data

Contains data transformation functions for the resvar model

train.py

This is the main training script which performs the following:

  1. Sets up the training environment and training parameters
  2. Sets up DataParallel if possible
  3. Performs unsupervised image reconstruction for the backbone and saves the model
  4. Performs superivsed map reconstruction and bounding box prediction, and saves the model.

About

Multi-Headed Variational Autoencoder using Residual Networks, for Self Driving

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published