Open tjasmin111 opened 11 months ago
@tjasmin111
Hi, thank you very much for your interest in my work.
I think there two ways to make the model smaller.
First, use smaller backbones. The code is this repository can be easily combined with any backbones similar to Resnet50. For example, to replace resnet50 with resnet18, simply change to resnet18 on line 145 of train_FGVC_Aircraft_ResNet50.py.
Second, use distillation. Actually, I have a new approach based on progressive learning and distillation, which has better performance and smaller parameters than this repository, and it will be released soon.
Interesting work. Is there a way to make the model smaller? How can I modify the model to do so? I need to increase the detection time to be fast.