Convolutional networks
The CNN era: deeper, wider and more modular convolutional backbones for vision.
| Concept | Introduced by | Years | Papers |
|---|---|---|---|
| Deconvnet feature visualisation Project activations back to pixel space to see what each CNN layer has learned. | ZFNet | 2013 | 0 |
| Inception module Parallel convolutions of several sizes in one block, to go deeper and wider at fixed cost. | GoogLeNet (Inception) | 2014–2021 | 12 |
| Very deep 3×3 ConvNets Stack many small 3×3 convolutions; depth alone drives accuracy. | VGG | 2014–2021 | 15 |
| Residual connection Add a layer's input to its output so very deep networks stay trainable. | ResNet | 2015–2022 | 32 |
| Cardinality (grouped convolutions) Split a block into many parallel same-shaped paths; the number of paths is a new scaling axis. | ResNeXt | 2016–2021 | 4 |
| Wide residual networks Shallower but wider residual blocks train faster and match very deep ResNets. | Wide ResNet | 2016 | 0 |