How are input images typically normalized for pretrained CNNs?
easyAnswer
- Subtract ImageNet channel means [0.485, 0.456, 0.406] and divide by stds [0.229, 0.224, 0.225] (RGB).
- This matches the distribution the model saw during pretraining — using different stats can hurt accuracy noticeably during transfer.
- When training from scratch, either compute stats on your dataset or normalize to [-1, 1] / [0, 1].
- Store the norm inside the model's preprocess step so it can't be forgotten in production.
Check yourself — multiple choice
- Never normalize
- Subtract ImageNet channel means/stds when using pretrained CNNs — matches pretraining distribution
- Only scale to [0, 255]
- Normalize the labels
Pretrained CNN → ImageNet mean/std normalization → matches training distribution.
#cnn#training
Practise Deep Learning
214 interview questions in this topic.