Meetings
- Discussed about ensemble and dropout uncertainty estimation
- Code for both was searched
- Asked to try the code
- Discussed about paper, idea about what and why question?
- Cleared evidential uncertainty error ie., replaced lenet model with resnet18
- Discussed about Deep learning models and decided to go with one model
- Calculated number of experiments to be done
- Discussed how visualization should look like ie., box-plots
To do
- Go through previous eccv and find papers about adversarial attack or patch attack
- We need to do qualitative analysis of the methods so need to learn about Cam,GradCam and (new method)
- Draw and add how visualization should look like
- Update what is done every 2 days through mail
- 30 days remaining to complete all experiments
- Discussed about visualization of both abstention class result and uncertainty estimation result
To do
- Could not discuss about models so one more meeting on 12 - 05 - 2022
- Discussed research questions
- Decided the publication which is ECCV for Adversarial Robustness
- Decided about the datasets which we will be working on namely, traffic sign, VGGface and Imagenette
To do
- Create table to see what experiments are remaining
- Think how to visualize the results
- Get ECCV paper format and add research questions to it
