Intellindust-AI-Lab/SURE ? reverse-engineered prompt

Build me a Python project that reproduces the SURE paper for reliable and robust deep networks.

I want to be able to train and test models for a few common problems like failure prediction, out of distribution detection, noisy labels, long tailed classification, and corruption robustness. Please make it work with datasets like CIFAR10, CIFAR100, Tiny ImageNet, ImageNet, Animal10N, Food101N, CIFAR LT, CIFAR10 C, and Stanford Cars, using the same kind of train, validation, and test flow described in the README. It should have simple command line scripts for training, fine tuning, validation, and testing, and save results in a CSV or similar file. If possible, include the common recipe options from the paper like baseline, mixup, CRL, SAM, and SWA so I can switch between them easily.

Set it up so I can create the environment, download or place the data in the right folders, run experiments from a single GPU, and inspect outputs without too much manual work. Look up current docs online if you need to.