architectureIntroduced by VGG · 2014
Very deep 3×3 ConvNets
Stack many small 3×3 convolutions; depth alone drives accuracy.
Drafted by AI · not yet reviewed
How this idea evolved
Suggested from citations. No curator has recorded what this idea builds on yet. These ideas from the same theme come from papers that VGG cites, directly or one step removed. Citation is a fact; the connection between the ideas is not verified.
Parallel convolutions of several sizes in one block, to go deeper and wider at fixed cost.
Project activations back to pixel space to see what each CNN layer has learned.
Papers using this
- 2015PReLU / He init
- 2015VQA
- 2015Flickr30k Entities
- 2015FM-IQA
- 2015Faster R-CNN
- 2015Inception v3
- 2015ResNet
- 2016Visual Genome
- 2016Wide ResNet
- 2016ResNeXt
- 2016Constrained beam search captioning
- 2016VQA v2
- 2018ImageNet-trained CNNs are biased towards
- 2019ImageNetV2
- 2021Swin