https://mer.dev/posts/conv2former-a-simple-transformer-style-convnet-for-visual-recognition/