Transformers, which are popular for language modeling, have been explored for solving vision tasks recently, e.g., the Vision Transformers (ViT) for image classification. The ViT model splits each image into a sequence of tokens with fixed length and then applies multiple Transformer layers to model their global relation for classification. However, ViT achieves inferior performance compared with CNNs when trained from scratch on a midsize dataset (e.g., ImageNet). We find it is because: 1) the simple tokenization of input images fails to model the important local structure (e.g., edges, lines) among neighboring pixels, leading to its low training sample efficiency; 2) the redundant attention backbone design of ViT leads to limited feature richness in fixed computation budgets and limited training samples.To overcome such limitations, we propose a new Tokens-To-Token Vision Transformers (T2T-ViT), which introduces 1) a layer-wise Tokens-to-Token (T2T) transformation to progressively structurize the image to tokens by recursively aggregating neighboring Tokens into one Token (Tokens-to-Token), such that local structure presented by surrounding tokens can be modeled and tokens length can be reduced; 2) an efficient backbone with a deep-narrow structure for vision transformers motivated by CNN architecture design after extensive study. Notably, T2T-ViT reduces the parameter counts and MACs of vanilla ViT by 200%, while achieving more than 2.5% improvement when trained from scratch on ImageNet. It also outperforms ResNets and achieves comparable performance with MobileNets by directly training on ImageNet. For example, T2T-ViT with ResNet50 comparable size can achieve 80.7% accuracy on ImageNet. 1

No abstract

Abstract-Recently, significant improvement has been made on semantic object segmentation due to the development of deep convolutional neural networks (DCNNs). Training such a DCNN usually relies on a large number of images with pixel-level segmentation masks, and annotating these images is very costly in terms of both finance and human effort. In this paper, we propose a simple to complex (STC) framework in which only image-level annotations are utilized to learn DCNNs for semantic segmentation. Specifically, we first train an initial segmentation network called Initial-DCNN with the saliency maps of simple images (i.e., those with a single category of major object(s) and clean background). These saliency maps can be automatically obtained by existing bottom-up salient object detection techniques, where no supervision information is needed. Then, a better network called Enhanced-DCNN is learned with supervision from the predicted segmentation masks of simple images based on the Initial-DCNN as well as the image-level annotations. Finally, more pixel-level segmentation masks of complex images (two or more categories of objects with cluttered background), which are inferred by using Enhanced-DCNN and image-level annotations, are utilized as the supervision information to learn the Powerful-DCNN for semantic segmentation. Our method utilizes 40K simple images from Flickr.com and 10K complex images from PASCAL VOC for step-wisely boosting the segmentation network. Extensive experimental results on PASCAL VOC 2012 segmentation benchmark well demonstrate the superiority of the proposed STC framework compared with other state-of-the-arts.

In natural images, information is conveyed at different frequencies where higher frequencies are usually encoded with fine details and lower frequencies are usually encoded with global structures. Similarly, the output feature maps of a convolution layer can also be seen as a mixture of information at different frequencies. In this work, we propose to factorize the mixed feature maps by their frequencies, and design a novel Octave Convolution (OctConv) operation 1 to store and process feature maps that vary spatially "slower" at a lower spatial resolution reducing both memory and computation cost. Unlike existing multi-scale methods, OctConv is formulated as a single, generic, plug-andplay convolutional unit that can be used as a direct replacement of (vanilla) convolutions without any adjustments in the network architecture. It is also orthogonal and complementary to methods that suggest better topologies or reduce channel-wise redundancy like group or depth-wise convolutions. We experimentally show that by simply replacing convolutions with OctConv, we can consistently boost accuracy for both image and video recognition tasks, while reducing memory and computational cost. An OctConv-equipped ResNet-152 can achieve 82.9% top-1 classification accuracy on ImageNet with merely 22.2 GFLOPs.