Deep learning models are often perceived as a “black box”, as it is difficult for us to understand how it works and why it works. Opening the black box is crucial for users to build trust with the technology. The speaker shall address the importance of feature visualization in convolutional neural networks and discuss various visualization approaches such as activation map, deconvolution, localization methods- Grad-CAM and saliency map. The speaker shall discuss practical implementation and showcase some results of visualization with natural images.