TTT++: When Does Self-Supervised Test-Time Training Fail or Thrive?

Authors: Yuejiang Liu, Parth Kothari, Bastien van Delft, Baptiste Bellot-Gurlet, Taylor Mordan, Alexandre Alahi

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically demonstrate that our modified version of test-time training, termed TTT++, outperforms state-of-the-art methods by significant margins on several benchmarks.
Researcher Affiliation Academia École Polytechnique Fédérale de Lausanne (EPFL){firstname.lastname}@epfl.ch
Pseudocode No The paper describes the proposed method in prose and through diagrams (Figure 2) but does not provide formal pseudocode or an algorithm block.
Open Source Code Yes Our code is available at https://github.com/vita-epfl/ttt-plus-plus.
Open Datasets Yes We train Res Net-50 [38] on CIFAR10/CIFAR100 [39] and test it on the CIFAR10-C/CIFAR100-C [1] datasets, which contain 15 types of algorithmically generated corruptions, such as noise, blur and snow effects. ... We finally validate the effectiveness of our method on the Vis DA-C dataset [42]
Dataset Splits No The paper discusses batch sizes and queue sizes for test-time adaptation but does not specify training/validation/test splits, only mentions that it trains on CIFAR10/CIFAR100 and tests on their corrupted versions.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments.
Software Dependencies No The paper mentions training models like ResNet-50 but does not specify any software dependencies or libraries with version numbers (e.g., PyTorch, TensorFlow, Python versions).
Experiment Setup Yes We use a batch size of 256 for test-time adaptation. In addition, we use a dynamic queue containing 16 batches of feature vectors for online feature alignment on CIFAR100-C.