Vgg

• @simonyanVeryDeepConvolutional2014
• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Deeper Alex Net
• Object detection and Image captioning
• 5x5 → two 3x3
• No of filters increase according to depth
• No of filters : increase by power of two
• Filter size : Odd numbers
• SGD + LR Schedule
• three non-linear activations (instead of one), which makes the function more discriminative
•