topdown

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

(★ 9)

Top-Down Networks

topdown 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Top-Down Networks
Code for reproducing the experiments of the paper [Top-Down Networks: A coarse-to-fine reimagination of CNNs](https://arxiv.org/abs/2004.07629). In case of any bugs/improvements, please reach Ioannis Lelekas ([giannislelekas@gmail.com](mailto:giannislelekas@gmail.com)).

## Requirements and Dependencies:
- Python >= 3.6
- NVIDIA GPU

We advise Anaconda for quick installation of all requirements and dependencies. Simply use the provided `requirements.yml` and `conda env create -f requirements.yml`.

## Organization:
This repository is organized as follows:
```
./                        #root directory of repository

data/                   #folder for downloading imagenette

notebooks/              #notebooks for visualizing results

src/                    #source code
    lib/                  #helper functions
    models/               #network architectures
    scripts/              #scripts for running the experiments

output/                 #output folder; generated after running the code
    adversarial/          #extracted adversarial attacks
    gradcam/              #gradcam heatmaps
    graphs/               #training curves
    history/              #training history (loss, acc, learning curve)
    models/               #models checkpoints
    output/               #generated output from training
    trained_weights/      #trained weights for models
```

## Usage:
All scripts for running the experiments are in `src/scripts/`. Command line inputs are given as comments within the scripts.

You may run the adversarial robustness (`run_adversarial_attacks.py`) and the localization experiment (`run_gradcam.py`) using models with pretrained weights.
To this goal, download ([link](https://drive.google.com/file/d/1cmO2sVblJ087ZQAEwQVgbyXUI2Gw96NQ/view?usp=sharing)) and extract `trained_weights.zip` and then place it under `src/output`, as shown in the repository organization section.

Otherwise, you need to train models from scratch using the corresponding scripts  (`src.scripts.run_BUvsTD.py` and `src.scripts.run_imagenette.py`).

The MD5 checksum of `trained_weights.zip`: 825009852499cb03781e45b905129615

## Citation:
For citing Top-Down Networks, please use the following:
```
@inproceedings{lelekas2020,
  title={Top-Down Networks: A coarse-to-fine reimagination of CNNs},
  author={I Lelekas and N Tomen and SL Pintea and JC van Gemert},
  booktitle={CVPR 2020 Workshop on Deep Vision},
  year={2020}
}
```

## Acknowledgements:
- https://github.com/BIGBALLON/cifar-10-cnn
- https://github.com/bethgelab/foolbox
- https://github.com/eclique/keras-gradcam
- https://github.com/koshian2/PCAColorAugmentation

topdown

topdown 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

topdown

topdown 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README