@@ -42,7 +42,7 @@ In addition to providing rich classification network structures and pretrained m
### knowledge distillation (SSLD)
Knowledge distillation refers to using the teacher model to guide the student model to learn specific tasks, ensuring that the small model has a relatively large effect improvement with the computation cost unchanged, and even obtains similar accuracy with the large model. PaddleClas provides a Simple Semi-supervised Label Distillation method (SSLD). With this method, different models on ImageNet1k-val dataset has 3% absolute improvement(Top1 accuracy) on ImageNet1k-val dataset. The following figure shows the models' performance after SSLD.
Knowledge distillation refers to using the teacher model to guide the student model to learn specific tasks, ensuring that the small model has a relatively large effect improvement with the computation cost unchanged, and even obtains similar accuracy with the large model. PaddleClas provides a Simple Semi-supervised Label Distillation method (SSLD). With this method, different models on ImageNet1k-val dataset have 3% absolute improvement(Top1 accuracy) on ImageNet1k-val dataset. The following figure shows the models' performance after SSLD.
