Code for my PhD thesis Characterization of a Reliability Domain for Image Classifiers.

Section 3.3 - Generate Synthetic Failure Cases

https://github.com/allglc/phd/tree/main/stylegan2

Section 3.4 - Generate Synthetic Uncertain Data

https://github.com/allglc/phd/tree/main/uncertainty-conditioned-gan

Section 3.5 - Discussion on Defining a Domain in the Latent Space

https://github.com/allglc/phd/tree/main/domain-toy
https://github.com/allglc/phd/tree/main/domain-images

Section 4.3 - Selective Classification Experiments

https://github.com/allglc/phd/tree/main/selective-classif

Section 4.4 - Making Confidence Calibration Methods More Data-Efficient

https://github.com/allglc/phd/tree/main/tva-calibration

Section 4.5 - Calibrate with Synthetic Data

https://github.com/allglc/phd/tree/main/calibration-with-synthetic-data

Section 5.3 - Textual Descriptions of Classifier Failures Using Text-to-Image Models & Section 5.4 - Using Textual Attributes to Define a Reliability Domain

https://github.com/allglc/phd/tree/main/semantics

Other

https://github.com/allglc/phd/tree/main/gan
https://github.com/allglc/phd/tree/main/diffusion
https://github.com/allglc/phd/tree/main/clip-confidence
https://github.com/allglc/phd/tree/main/clip-characterize-classifier
https://github.com/allglc/phd/tree/main/dream-domain