The repository contains the code that I wrote and the experients that I conducted for my dissertation with the title Improving Convolutional Neural Network-based Image Classification by Exploiting Network Layer Information. The dependencies of the code can be found in each Jupyter notebook.
The directory patch_init contains the code and experiments for my method Image Patch-based Model Weight Initialization in order to make model training more robust with respect to the choice of the learning rate parameter. This method is described in chapter 4 of my dissertation and in my corresponding publication:
@inproceedings{lehmann/ebner:2021a,
author = {Lehmann, Daniel and Ebner, Marc},
title = {{Are Image Patches Beneficial} for {Initializing Convolutional Neural Network Models}?},
booktitle = {{Proceedings} of the 16th {International Joint Conference} on {Computer Vision}, {Imaging} and {Computer Graphics Theory} and {Applications} ({VISIGRAPP} 2021) - {Volume} 5: {VISAPP}},
year = {2021},
pages = {346-353},
publisher = {SciTePress},
organization = {INSTICC},
}
If you use my code, then please reference the corresponding paper according to the bibtex above.
The directory scus contains the code and experiments for my method Subclass-based Undersampling in order to address model training using a class-imbalanced training dataset. This method is described in chapter 5 of my dissertation and in my corresponding publication:
@inproceedings{lehmann/ebner:2022a,
author = {Lehmann, Daniel and Ebner, Marc},
title = {{Subclass-based Undersampling} for {Class-imbalanced Image Classification}},
booktitle = {{Proceedings} of the 17th {International Joint Conference} on {Computer Vision}, {Imaging} and {Computer Graphics Theory} and {Applications} ({VISIGRAPP} 2022) - {Volume} 5: {VISAPP}},
year = {2022},
pages = {493-500},
publisher = {SciTePress},
organization = {INSTICC},
}
If you use my code, then please reference the corresponding paper according to the bibtex above.
The directory laca contains the code and experiments for the naive version of my method Layer-wise Activation Cluster Analysis in order to detect out-of-distribution samples at inference with respect to a trained model. This method is described in chapter 6 of my dissertation (especially in section 6.2.3) and in my corresponding publication:
@inproceedings{lehmann/ebner:2021b,
author = {Lehmann, Daniel and Ebner, Marc},
title = {{Layer-Wise Activation Cluster Analysis} of {CNN}s to {Detect Out-of-Distribution Samples}},
booktitle = {{Proceedings} of the 30th {International Conference} on {Artificial Neural Networks} and {Machine Learning} - {ICANN} 2021},
year = {2021},
editor = {Farka{\v{s}}, Igor and Masulli, Paolo and Otte, Sebastian and Wermter, Stefan},
series = {Lecture Notes in Computer Science},
pages = {214-226},
address = {Cham, Switzerland},
publisher = {Springer},
}
If you use my code, then please reference the corresponding paper according to the bibtex above.
The directory laca2 contains the code and experiments for the more sophisticated version of my method Layer-wise Activation Cluster Analysis in order to detect out-of-distribution samples at inference with respect to a trained model. This method is described in chapter 6 of my dissertation (especially in section 6.2.4) and in my corresponding publication:
@inproceedings{lehmann/ebner:2022b,
author = {Lehmann, Daniel and Ebner, Marc},
title = {{Calculating} the {Credibility} of {Test Samples} at {Inference} by a {Layer-wise Activation Cluster Analysis} of {Convolutional Neural Networks}},
booktitle = {{Proceedings} of the 3rd {International Conference} on {Deep Learning Theory} and {Applications} - {Volume} 1: {DeLTA}},
year = {2022},
pages = {34-43},
address = {Lisbon, Portugal},
publisher = {SciTePress},
organization = {INSTICC},
}
If you use my code, then please reference the corresponding paper according to the bibtex above.
The directory laca3 contains additional experiments for the more sophisticated version of my method Layer-wise Activation Cluster Analysis (see laca2). These experiments are described in section 6.3.4 and 6.3.5 of chapter 6 of my dissertation and in my corresponding publication:
@inproceedings{lehmann/ebner:2023,
author = {Lehmann, Daniel and Ebner, Marc},
title = {{Reliable Classification} of {Images} by {Calculating} their {Credibility Using} a {Layer-wise Activation Cluster Analysis} of {CNN}s},
booktitle = {{Deep Learning Theory} and {Applications} - {DeLTA} 2022 ({Revised Selected Papers})},
year = {2023},
editor = {Fred, Ana and Sansone, Carlo and Gusikhin, Oleg and Madani, Kurosh},
series = {Communications in Computer and Information Science},
pages = {33-55},
address = {Cham, Switzerland},
publisher = {Springer},
}
If you use my code, then please reference the corresponding paper according to the bibtex above.
Coming soon.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent