how to distill a student with specific structure like "16_16_32_16_32_32_16_16"?

mit-han-lab / gan-compression

[CVPR 2020] GAN Compression: Efficient Architectures for Interactive Conditional GANs

Other

1.1k stars 148 forks source link

how to distill a student with specific structure like "16_16_32_16_32_32_16_16"? #14

Closed MarSaKi closed 4 years ago

MarSaKi commented 4 years ago

I notice your code can support student with structure "16_16_16...._16", "24_24..._24", "3232..._32", but how to distill a student like "16_16_32_16_32_32_16_16"? If I add config_str in distill options, will it work?

lmxyy commented 4 years ago

Such distillation is somewhat like fine-tuning and can be done with train_supernet.py.

You could refer to the script finetune.sh to see how to fine-tune or distill a customized structure with our code.

MarSaKi commented 4 years ago

Thanks, I want to know your experiment in fig 6, is "pruning+distill" means?: teacher: "64_64_64_64_64_64_64_64" student: "32_32_32_32_32_32_32_32" "24_24_24_24_24_24_24_24" "16_16_16_16_16_16_16_16" for each point?