gaopengpjlab

Results 63 comments of gaopengpjlab

Can you give us guide how to run your downloaded convmae.pth model to finetune?

Please refer to the following guidance for ConvMAE fine-tuning. https://github.com/Alpha-VL/ConvMAE/blob/main/FINETUNE.md

Hello, how to finetune own datasets

Please follow the dataset format of Imagenet. Basically, you can need to build two folders named dog and cat then put images into the corresponding folder.

Hello, how to finetune own datasets

train ----dog ----cat val ----dog ----cat The folder structure is similar to the above illustration.

about pretrain model convmae.pth

Thanks for your interest in ConvMAE. The checkpoint only contain encoder weight as only encoder weight is required for downstream tasks.

about pretrain model convmae.pth

We will find a way to share the full model. Stay tuned.

ConvMAE flops for classification ?

We will update the FLOPs of classification in a few days. Please stay tuned.

How to pretrain CNN backbones like ResNet and EfficientNet?

This is a great research problem that is beyond the scope of ConvMAE. We will explore the pretraining of CNN backbones in the future.

When to release other model weights rather than base model?

Thanks for your interest. We are working on a new version of ConvMAE with faster pretraining and improved representation ability. Pretrained models ranging from small to huge shall be released....

When to release other model weights rather than base model?

We update the pretrained weights of ConvMAE V2 small, base, large and huge in the README.

When to release other model weights rather than base model?

We will update the details of ConvMAE V2 later.