Backto DeepCV
拿过来一个 pre-trained mode,希望能 fine-tune 成解决当前问题的 model。又是一门手艺。
首先,定策略
动手前要先分析燃油(new dataset)的数量和质量(与pre-used 的dataset 的相似性)。
x | very similar dataset | very different dataset |
---|---|---|
very little data | train a linear classifier on the features extracted from the top layers of pre-trained models. | Trouble Warning. Since the data is limited, it seems better to only train a linear classifier. |
quite a lot of data | fine-tune a few top layers of pre-trained models with a small learning rate | a large number of layers should be fine-tuned on your data also with a small learning rate for improving performance |
然后,精调
精调