pytorch-image-models

pytorch-image-models copied to clipboard

When instantiating `torchvision`'s `ImageNet` or `ImageFolder`, the `download` argument is passed, even if these two classes do not take this argument. This PR removes the argument from the `torch_kwargs` dict...

dedeswim

Hi, we are a group of engineers from Bytedance Inc. This year, our team published the work: "Next-ViT: Next Generation Vision Transformer for Efficient Deployment in Realistic Industrial Scenarios"(https://arxiv.org/abs/2207.05501) (https://github.com/bytedance/Next-ViT)....

XiaXin-Aloys

[BUG] Fail to turn swin_base_patch4_window12_384 into feature extractor

1

Hi, I'm trying to use `swin_base_patch4_window12_384` model to extract features, but I meet some error as follows: ```python model = timm.create_model("swin_base_patch4_window12_384", features_only=True, pretrained=False) ``` ```bash AttributeError: 'SwinTransformer' object has no...

rentainhe

The behavior is not obvious to me. Perhaps it's useful to mention this here to avoid confusion.

F5D7

Do we have plan to support MobileViT v3? MobileViTv3: Mobile-Friendly Vision Transformer with Simple and Effective Fusion of Local, Global and Input Features https://arxiv.org/abs/2209.15159 https://github.com/micronDLA/MobileViTv3 

Amadeus-AI

Add DaViT

12

Adapt the DaViT model from https://arxiv.org/abs/2204.03645 and https://github.com/dingmyu/davit. Notably, the model performs on par with many new models, such as MaxViT, whilst having higher throughput, a design that should allow...

fffffgggg54

[BUG] Not able to load `eva_giant_patch14_560.m30m_ft_in22k_in1k` model

1

I think there is a `split` function in the reading of the model name that doesn't read the full name of the model. ## To Reproduce ```python import timm model...

MohamedAliRashad

Add FocalNet arch, refactor Swin V1/V2 for better feature extraction and HF hub multi-weight support

1

Since FocalNet and Swin related (and both need refactoring for better feat extraction support), combining * Introduction of FocalNet arch * Refactor Swin V1/V2, possibly other similar arch that could...

rwightman

What batch size number other than batch size of 1024 have you tried when training a DeiT or ViT model? In the paper, DeiT (https://arxiv.org/abs/2012.12877), they used a batch size...

Phuoc-Hoan-Le

Hello, Vision transformers in timm currently use a custom implementation of attention instead of `nn.MultiheadAttention`. Pytorch 2.0 will come with [flash attention](https://arxiv.org/abs/2205.14135) which is an exact implementation of attention, but...

SimJeg