Skip to content

Latest commit

 

History

History
34 lines (23 loc) · 2.43 KB

review-todo.md

File metadata and controls

34 lines (23 loc) · 2.43 KB
Raw

3.3-3.11: What do CMR and CNMR mean? Maybe explain shortly in 3.2 (write out the abbreviation once).

  • Interpretation: „Larger networks suffer from a complex set of layers and nodes that come with no benefits and even weakens the purpose as the increased calculation time brings inconvenience to the potential user.“

    • -> Do you mean that „globally" or only for your investigated case of fingerprints? You can then express it a little better here so that it doesn't sound like your interpretation applies to every research case.
    • The author does answer all the questions, even though the questions have to be written out in the abstract and introduction section, to better relate.

  • The references are missing both in the paragraphs themselves and in the list of references or written out in the middle of a sentence.

  • Try to put the cites at the end of a paragraph, to not disturb the reading flow.

  • Section 2.4 - Methodology: You don't need to force a fixed threshold. You could just plot the DET for all possible thresholds and present a visual comparison.

  • DONE Use standardised vocabulary, e.g.:

    • "liveness score" -> "PAD score".
    • "presentation attack" -> "attack presentation"
    • CMR/CNMR are neither defined nor commonly used for biometric PAD evaluations
    • align "true positive/negate rate" to the PAD case with APCER and BPCER

  • Do not mix the terms training, validation, and testing (Section 2.1 Dataset).

    • While it is possible to call the testing 'validation', validation itselfs refers to a particular part of the training process. In deep learning, you can train the model on the training set and directly evaluate the performance on the validation set. When the validation loss is low enough, you consider the training done, and use the test set to report the performance on unseen data.

  • NOIMP maybe a litle more information about the used algorithms would be interesting and help to differentiate between them

  • DONE? maybe a little bit more information about the topic in general would be interesting, there is no information about the importance of the topic in the abstract

  • NOIMP you may also need to explain the chosen representaions (eg liquid ecoflex)

  • sentences like which... or that... are used quite often too and may be replaced like this: which also took the longest to train ==> also taking the longest to train this may help to shorten some sentences and makes it more easy to follow