cdf_att_classification

Classes = {0: 'cat', 1: 'dog', 2: 'flower'}

In this project we use both Resnet and Self-attention layer for cdf-Classification. Specifically, For Resnet, we extract low level features from Convolutional Neural Network (CNN) trained on Dogcatflower_2 dataset(details show later).
We take inspiration from the Self-attention mechanism which is a prominent method in cv domain. We also use Grad-CAM algorithm to Visualize the gradient of the back propagation of the pretrain model to understand this network. The code is released for academic research use only. For commercial use, please contact [lailanqing_charlie@163.com].

Installation

Clone this repo.

git clone https://github.com/Alan-lab/cdf_classification
cd cdf_classification/

This code requires pytorch, python3.7, cv2, d2l. Please install it.

Dataset Preparation

For cdf_classification, the datasets must be downloaded beforehand. Please download them on the respective webpages. Please cite them if you use the data.

Preparing Cat and Dog Dataset. The dataset can be downloaded here.

Preparing flower Dataset. The dataset can be downloaded here.

You can also download Dogcatflower_2 dataset(made from above datasets) use the following link:

Link:https://pan.baidu.com/s/1ZcP_isbbRQBq9BHU6p_VtQ

key:oz7z

Training New Models

1. Prepare your own dataset like this (https://github.com/Alan-lab/data/Dogcatflower_2).

2. Training:

python main.py

model.pth will be extrated in the folder ./cdf_classification.

If aver_test_acc < 0.75, model.pth will not save(d2l.train_ch6).

3.Predict

Prepare your valid dataset like this (https://github.com/Alan-lab/data/catsdogsflowers/valid1).

python Predict/predict.py

4.Class Activation Map The response size of the feature map is mapped to the original image, allowing readers to understand the effect of the model more intuitively. Prepare your picture like this (https://github.com/Alan-lab/data/Dogcatflower/test/flower/flower.1501.jpg).

python Viewer/Grad_CAM.py

5. More details can be found in folder.

The Experimental Result

1. Preformance

dataset	Cat-acc	Dog-acc	flower-acc
Dogcatflower_2_train	96.2	88.7	93.6
Dogcatflower_2_test	72.7	69.2	89.7
catsdogsflowers_valid1	75.1	76.9	91.4
catsdogsflowers_valid2	75.5	73.5	92.9

2.Visualization

Postive sample

Negative sample

Multi-attention

Acknowledgments

This work is mainly supported by (https://courses.d2l.ai/zh-v2/) and CSDN.

Contributions

If you have any questions/comments/bug reports, feel free to open a github issue or pull a request or e-mail to the author Lailanqing (lailanqing_charlie@163.com).