ConvNeXt

A family of purely convolutional networksconvolutional neural network, n.: A kind of neural network built for images and other grid-like data. It slides small pattern detectors across the input, finding edges first, then shapes, then whole objects. reached by taking a standard ResNetResNet, prop. n.: A family of image-recognition networks from 2015 whose shortcut connections allowed far deeper networks than before, over 100 layers, and became a standard building block across AI. and changing it one design decision at a time to resemble a vision transformervision transformer, n.: A way of applying the transformer, the design behind chatbots, to pictures: the image is cut into small square tiles that the model reads like words in a sentence., while keeping every component convolutional: a 'patchify' stem (4 × 4 convolution with stridestride, n.: How far a scanning window jumps each time it moves across an image. Jumping two pixels instead of one halves the size of the result. 4), transformer-like stage proportions, large 7 × 7 depthwise convolutionsdepthwise separable convolution, n.: A cheaper way to do the main operation of image networks: first scan each channel of the input separately, then mix the channels together with a tiny one-by-one step., inverted bottlenecks, fewer activation and normalization layers, GELUGELU, abbr.: A smooth version of the ReLU rule, used in models such as BERT and GPT: it keeps large positive values, removes large negative ones, and blends softly in between. instead of ReLUReLU, abbr.: The most common rule a neuron uses to produce its output: keep the number if it is positive, replace it with zero if it is negative. Simple, fast and surprisingly effective., layer normalizationlayer normalization, n.: A layer that rescales the numbers inside a network, separately for each example, so they keep a steady average and spread. It, or a simpler variant, is a standard part of the models behind today's chatbots. instead of batch normalizationbatch normalization, n.: A layer that rescales the numbers flowing through a network so that, across each batch of training examples, they keep a steady average and spread. It makes deep networks train faster and more reliably., and transformer-style training recipes. With large-scale pretraining it scored 87.8% top-1 accuracy on ImageNet and beat the Swin Transformer at COCO object detection and ADE20K segmentation. Its lesson, that much of the transformers' advantage in vision came from training recipes and design details rather than attention itself, fed a wider debate about architecture versus scale; ConvNeXt V2 (2023) added masked-autoencoder pretraining.

On our edge GPU, ConvNeXt-T matched the vision transformer's accuracy at lower latency.