search

leondgarse / keras_efficientnet_v2

self defined efficientnetV2 according to official version. Including converted ImageNet/21K/21k-ft1k weights.
Apache License 2.0
78 stars 19 forks source link

Dead link to model "Imagenet" for EffV2XL #13

Closed DennisMedved closed 2 years ago

DennisMedved commented 3 years ago

Thanks for the nice job on implementing the EffNet-models!

I tried to use the EffV2XL model; but the link to the model "Imagenet" doesn't not seem to work.

https://github.com/leondgarse/keras_efficientnet_v2/releases/download/effnetv2_pretrained/efficientnetv2-xl-imagenet.h5

It doesn't seem to be in the release: https://github.com/leondgarse/keras_efficientnet_v2/releases/

Best regards, Dennis

leondgarse commented 3 years ago

Right, we don't have the EffV2XL model on imagenet, as it's not provided from official. I forgot that link.. Just updated and set default pretrained="imagenet21k-ft1k" for EffV2XL.

DennisMedved commented 3 years ago

Thank you! :)

What is the difference between "ImageNet21K" and "Imagenet21k-ft1k"?

I could not find any documentation on it.

leondgarse commented 3 years ago

ImageNet21K is the raw model pre-trained on ImageNet21K dataset. The expected output num_classes should be 21843. Imagenet21k-ft1k means the ImageNet21K model fine-tuned on imagenet, and its expected output num_classes is 1000, same with imagenet model. It's also called ImageNet22K somewhere.