Skip to content

Sprint Goals

Jump to bottom
Kaushik Manjunatha edited this page Jan 7, 2024 · 27 revisions

Research Goals

  • Survey on the different Uncertainty Estimation methodologies for Surface Normal Estimation
  • Comparing the benchmark loss function (von MisesFisher distribution) with the wrapped nrml and wrapped cauchy loss functions

Sprint Goals

December 15th 2023

  • 1. Write the Metrics and the Loss functions used part of the Thesis report.
  • 2. Fixing the training pipeline wrt python3.8 to make it run in neptune logger
  • 3. Add Neptune logging to the whole pipeline by migrating it to python3.8
  • 4. Getting the Dataset Trained and Tested on the vonMF. (Fixed the error in the DDP and now the training and the testing works fine)

November 15th 2023

  • 1. Write a blog about UE and the 3 value representation SNE.
  • 2. Understanding the architecture, the representation, and the outputs and blogging it. (Conversion from Mu K to 3 values).
  • [-] 3. Getting the Dataset Trained and Tested on the vonMF. (Partially done due to error in DDP, No Cluster Availability)
  • 4. Finding a toy dataset is required for the same network with the same output layer and loss functions.
  • 5. Write the Introduction part of the Thesis report.

October 15th 2023

  • 1. Register the Thesis
  • 1.1. Expand the related works, add dataset information
  • 2. Toy dataset - S03 learning
  • 3. Write the Dataset part of the Thesis Report

September 15th 2023

  • 1. Write Thesis Proposal
  • 1.1 First draft of proposal is ready. Need to get the approval from the supervisors to register.
  • 2. Register the Thesis
  1. Understanding the benchmark paper, write a blog, Forward snowballing of the benchmark paper
  • 3.1.Write a blog - a small blog on the benckmark paper is done.
  • 3.2.Forward snowballing of the benchmark paper. - a excel sheet with all the information about the paper and its metrics along with dataset is done.
  1. Do a Survey on relevant Datasets and Metrics
  • 4.1. Create an Excel sheet + one-page summary of all the loss functions, metrics - Excel sheet and one page summary of loss functions and metrics is done.
  • 4.2. If a single dataset has been used, write all the metrics related to the paper - Metrics of the related paper is noted down.
  • 5. Run the toy dataset on the collab and test the Deebuls loss function.
  • Problem with the challenge - needs premium account, cannot fetch the dataset.
  • 5.2 SO3 learning -Reading the paper and the generation the simulation data.
  • 6. add Neptune logging Plot wrapped Cauchy and wrapped normal functions in Python and compare with the VMF - Neptune, Weights and Biases has version problems with the current version of the python used.

To Do

  • Update ReadMe
Clone this wiki locally