zju-vipa / KamalEngine

Knowledge Amalgamation Engine
Apache License 2.0
97 stars 18 forks source link

KA for VIT #43

Closed maxwZJU closed 10 months ago

maxwZJU commented 10 months ago

Provide a comprehensive and robust pipeline and a data preprocessing method that allows training with own datasets. Implement KA on vision transformers to accomplish image classification tasks.(finetuned on several datasets such as Stanford Dogs) Provide a multimodal transfer method to apply KA for audio classification.