What is ConvNeXt's philosophy?
hardAnswer
- Liu et al. (2022) 'modernize' a ResNet using tricks from Swin: bigger kernels (7×7 depthwise), GELU + LayerNorm, inverted bottlenecks, fewer activations / norms, and modern training recipes (AdamW + Mixup + RandAugment).
- The result is a pure-conv architecture that matches Swin Transformer on ImageNet, detection, and segmentation.
- Shows attention isn't magical — modern training + design does most of the work.
Check yourself — multiple choice
- ConvNeXt uses global attention
- Modernized ResNet: 7x7 depthwise convs, GELU, LN, modern training — matches Swin without attention
- Only for text
- ConvNeXt has no convolutions
ConvNeXt: bring Swin-era design tricks into a pure CNN → matches transformers.
#architectures#computer-vision#cnn
Practise Deep Learning
214 interview questions in this topic.