Swin-Transformer

Swin-Transformer copied to clipboard

i want to know about data set, how enlarge the ImageNet-22K datasets by 5 times to reach million images with noisy labels? just data augmentation? hoping for your replation~ thank...

peiyingxin

Small changes in swin_transformer.py for Image size scale

3

Thanks for your remarkable work, when I use the pretrained model ，I found it difficult to scale the input Image size，even I use the image size like `window_size*2^n` , it...

zichengsaber

How can I get the model score on single image to classify the image as real or fake Thanks

sathyagorla

ape(absolute positional embedding) is set False by default, is it OK?

3

The parameter `ape`, which stands for adding absolute position embedding to the patch embedding, is set `False` by default in the config file. To my knowledge, for transformer models, input...

feiyangsuo

Minor discrepency between training log reported accuracy and evaluation accuracy on ImageNet.

1

Thanks for releasing the code! However, we use your codebase for training and testing some models and evaluate their performance on ImageNet1K(w/o pretrained on ImageNet22K). However, according to the training...

FreddieRao

How to finetune the 384 models with window size 12?

5

Great works! I have some questions about fine-tuning on ImageNet 1K. In the paper, you claimed 384^2 input models are obtained by fine-tuning as also pointed by #24: >For other...

silent-chen

can you provide the parameter settings for finetune on 384 from 224 of image classification?

2

can you provide the parameter setting for finetune on 384 from 224?

Dongshengjiang

Could you please provide the training schedule for imagenet22k pretrain?

1

The current configs seem to be containing the fine-tuning process from imagenet22k pretrained to imagenet 1k only. Could you please provide the imagenet22k pretrain schedule?

zbwxp

How to Fine-tune ImageNet-1K by ImageNet-21K pretrained

2

Thanks for you great work. I want to fine-tune on ImageNet-1K by ImageNet-21K pretrained bacause it's more faster to reproduce your result. So how to do it? I try to...

tiancity-NJU

apex

1

hnjzbss