This week we will not continue the deep clustering we will explore the
1.Attention mechanisms
2.Making the new dataset
3.Feature concatenation techniques
1. Attention mechanism
The class activation tecnhnique suffers through a limitation that it requires feature maps to directly precede the softmax layers, so it is applicable to a particular kind of CNN architectures that perform global average pooling over convolutional maps immediately before prediction. Grad-Cam, unlike CAM, uses the gradient information flowing into the last convolutional layer of the CNN to understand each neuron for a decision of interest. Our idea is to use the pretrained networks such as vgg, resnet , inception , densenet and etc and see the attention it is making on the input images
The code can be found on here
https://github.com/Himani2000/GSOC_2020/blob/master/Image_clustering/GradCAM.ipynb

Inference from the code :
2. Making the new dataset
What does making a new dataset means ? So our dataset is a screenshots of the multimedia which mostly contains people in them since we saw in the attention method that the focus is not on the people but is rather in some other parts of the image , now to reduce the noise and make the attention more focused towards the people we make two new datasets A and B.
where A is the dataset of detected persons
and B is the dataset of the detected faces plus some surrounding area
Results for the faces dataset
canny_ideology_face_convolutional1-hierarchical-pca-tsne_agg_labels.pdf
ideology_face_model_mobilenet-density-pca_90_hdbscan_labels.pdf
ideology_face_model_mobilenet-density-pca_hdbscan_labels.pdf
ideology_face_dataset_results_hog_-density-tsne_dbscan_labels.pdf
Results for the person dataset
ideology_person_model_mobilenet-density-pca-tsne_dbscan_labels.pdf
ideology_person_model_resnet50-density-pca_90_birch_labels.pdf
grayscale_ideology_person_convolutional2-density-tsne_dbscan_labels.pdf


The code (jupyter notebook can be found on here , please see the jupyter notebook if you are interested in the visualization ) for the person detection
https://github.com/Himani2000/GSOC_2020/blob/master/Data_extraction/Images/person-detection.ipynb The python code
https://github.com/Himani2000/GSOC_2020/blob/master/Data_extraction/Images/person_detecton.py The code(jupyter notebook ) for face detection with some surrounding area
https://github.com/Himani2000/GSOC_2020/blob/master/Data_extraction/Images/Face-Detection.ipynb The code (python code) for the face detection with some surrounding area
https://github.com/Himani2000/GSOC_2020/blob/master/Data_extraction/Images/face_detect.py See you next week