This repository hosts the code and data lists for our learning-based eXplainable AI (XAI) method called T-TAME, for Convolutional and Transformer-like Deep Neural Network-based (DNN) image classifiers. Our method receives as input an image and a class label and produces as output the image regions that the DNN has focused on in order to infer this class. T-TAME uses an attention mechanism (AM), trained end-to-end along with the original, already-trained (frozen) DNN, to derive class activation maps from feature map sets extracted from selected layers. During training, the generated attention maps of the AM are applied to the inputs. The AM weights are updated by applying backpropagation on a multi-objective loss function to optimize the appearance of the attention maps (minimize high-frequency variation and attention mask area) and minimize the cross-entropy loss. This process forces the AM to learn the image regions responsible for the DNN’s output. Two widely used evaluation metrics, Increase in Confidence (IC) and Average Drop (AD), are used for evaluation. Additionally, the promising ROAD framework is also used for evaluation. We evaluate T-TAME on the ImageNet dataset, using the VGG16, ResNet-50, and ViT-B-16 DNNs. Our method outperforms the state-of-the-art methods in terms of IC and AD and achieves competitive results in terms of ROAD. We also provide a detailed ablation study to demonstrate the effectiveness of our method.
- This repository contains the code for training, evaluating, and applying T-TAME, using VGG-16, ResNet-50, or ViT-B-16 as the pre-trained backbone network along with the Attention Mechanism and our selected loss function. There is also a guide on applying TAME to any DNN image classifier.
- It also contains the trained T-TAME attention mechanism for VGG-16, ViT-B-16, and ResNet-50 and the L-CAM method for the VGG-16 and ResNet-50 classifiers, used for comparisons. The checkpoints are bundled in a submodule repository located at HuggingFace hub (link) using
git-lfs. - In
T-TAME/datalist/ILSVRC, text files with annotations for 2000 randomly selected images to be used at the validation stage (Validation_2000.txt) and 2000 randomly selected images (exclusive of the previous 2000) for the evaluation stage (Evaluation_2000.txt) of the L-CAM methods. - The ILSVRC 2012 dataset images should be downloaded by the user manually.
Make sure that you have a working git, git-lfs, Python 3, cuda, and poetry installation before proceeding.
-
To install git, follow the instructions here.
-
To install python, follow the instructions here.
-
To install cuda, follow the instructions here.
-
To install poetry, follow the instructions here.
-
To install git-lfs, follow the instructions here
-
Clone this repository:
git lfs install git clone --recurse-submodules git@github.com:IDT-ITI/T-TAME.git
-
Go to the locally saved repository path:
cd T-TAME -
Install:
poetry install
-
Create a
.envfile inpl_scripts. The file should contain the following key-value pair:DATA=<path to imagenet dataset> LIST=<path to datalist folder>
The
LISTpath should be the path to the./datalist/ILSVRCfolder if using the ImageNet dataset.
Note: You may need to modify the venv activate script in the case that cuda is already installed on your machine. If so, add this line:
export LD_LIBRARY_PATH=.../venv/lib/python3.8/site-packages/nvidia/cublas/lib/:$LD_LIBRARY_PATH
You can evaluate the compared methods, including T-TAME with the {vgg16, resnet50, vit_b_16} backbones using the following command:
python pl_scripts/{vgg16, resnet50, vit_b_16}_comparisons.pyYou can train the T-TAME method using the following command:
python pl_scripts/{vgg16, resnet50, vit}_TAME.pyYou can generate explanation maps for the T-TAME method using the following command:
python pl_scripts/TAME_print_mask.pyYou can generate explanation maps for the compared methods using the following command:
python pl_scripts/other_methods_print_mask.pyIf you find our T-TAME method, code, or pretrained models useful in your work, please cite the following publication:
- M. V. Ntrougkas, N. Gkalelis, and V. Mezaris, “T-TAME: Trainable Attention Mechanism for Explaining Convolutional Networks and Vision Transformers.”, IEEE Access, 2024. doi: 10.1109/ACCESS.2024.3405788.
@ARTICLE{10539635,
author={Ntrougkas, Mariano V. and Gkalelis, Nikolaos and Mezaris, Vasileios},
journal={IEEE Access},
title={T-TAME: Trainable Attention Mechanism for Explaining Convolutional Networks and Vision Transformers},
year={2024},
volume={},
number={},
pages={1-1},
keywords={Convolutional neural networks;Transformers;Task analysis;Computer architecture;Image classification;Computational modeling;Training;CNN;Vision Transformer;Deep Learning;Explainable AI;Model Interpretability;Attention},
doi={10.1109/ACCESS.2024.3405788}}You may want to also consult and, if you find it useful, also cite our earlier works on this topic (methods TAME, L-CAM-Img, L-CAM-Fm):
- M. Ntrougkas, N. Gkalelis and V. Mezaris, "TAME: Attention Mechanism Based Feature Fusion for Generating Explanation Maps of Convolutional Neural Networks," in 2022 IEEE International Symposium on Multimedia (ISM), Italy, 2022 pp. 58-65. doi: 10.1109/ISM55400.2022.00014
- Gkartzonika, I., Gkalelis, N., Mezaris, V. (2023). Learning Visual Explanations for DCNN-Based Image Classifiers Using an Attention Mechanism. In: Karlinsky, L., Michaeli, T., Nishino, K. (eds) Computer Vision – ECCV 2022 Workshops. ECCV 2022. Lecture Notes in Computer Science, vol 13808. Springer, Cham. https://doi.org/10.1007/978-3-031-25085-9_23
@INPROCEEDINGS{10019620,
author={Ntrougkas, Mariano and Gkalelis, Nikolaos and Mezaris, Vasileios},
booktitle={2022 IEEE International Symposium on Multimedia (ISM)},
title={TAME: Attention Mechanism Based Feature Fusion for Generating Explanation Maps of Convolutional Neural Networks},
year={2022},
volume={},
number={},
pages={58-65},
keywords={Training;Visualization;Computational modeling;Neural networks;Computer architecture;Streaming media;Feature extraction;CNNs;Deep Learning;Explainable AI;Interpretable ML;Attention},
doi={10.1109/ISM55400.2022.00014}}
@InProceedings{10.1007/978-3-031-25085-9_23,
author="Gkartzonika, Ioanna
and Gkalelis, Nikolaos
and Mezaris, Vasileios",
editor="Karlinsky, Leonid
and Michaeli, Tomer
and Nishino, Ko",
title="Learning Visual Explanations for DCNN-Based Image Classifiers Using an Attention Mechanism",
booktitle="Computer Vision -- ECCV 2022 Workshops",
year="2023",
publisher="Springer Nature Switzerland",
address="Cham",
pages="396--411",
isbn="978-3-031-25085-9"
}Copyright (c) 2024, Mariano Ntrougkas, Nikolaos Gkalelis, Vasileios Mezaris / CERTH-ITI. All rights reserved. This code is provided for academic, non-commercial use only. Please also check for any restrictions applied in the code parts and datasets used here from other sources. For the materials not covered by any such restrictions, redistribution and use in source and binary forms, with or without modification, are permitted for academic non-commercial use provided that the following conditions are met:
Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation provided with the distribution.
This software is provided by the authors "as is" and any express or implied warranties, including, but not limited to, the implied warranties of merchantability and fitness for a particular purpose are disclaimed. In no event shall the authors be liable for any direct, indirect, incidental, special, exemplary, or consequential damages (including, but not limited to, procurement of substitute goods or services; loss of use, data, or profits; or business interruption) however caused and on any theory of liability, whether in contract, strict liability, or tort (including negligence or otherwise) arising in any way out of the use of this software, even if advised of the possibility of such damage.
The T-TAME implementation was built in part on code previously released in the TAME repository.
The code for the methods that are used for comparison is taken from the TAME repository for TAME, the L-CAM repository for L-CAM-Img, the RISE repository for RISE, the IIA repository for IIA, the Transformer-Explainability repository for the Transformer LRP method and the pytorch-gradcam repository for all of the remaining utilized methods.