AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation

• URL: http://arxiv.org/abs/2411.04967v1
• Date: Thu, 07 Nov 2024 18:43:17 GMT
• Title: AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation
• Authors: Anil Kag, Huseyin Coskun, Jierun Chen, Junli Cao, Willi Menapace, Aliaksandr Siarohin, Sergey Tulyakov, Jian Ren,
• Abstract summary: We introduce AsCAN -- a hybrid architecture, combining both convolutional and transformer blocks. AsCAN supports a variety of tasks: recognition, segmentation, class-conditional image generation. We then scale the same architecture to solve a large-scale text-to-image task and show state-of-the-art performance.
• Score: 48.82264764771652
• License:
• Abstract: Neural network architecture design requires making many crucial decisions. The common desiderata is that similar decisions, with little modifications, can be reused in a variety of tasks and applications. To satisfy that, architectures must provide promising latency and performance trade-offs, support a variety of tasks, scale efficiently with respect to the amounts of data and compute, leverage available data from other tasks, and efficiently support various hardware. To this end, we introduce AsCAN -- a hybrid architecture, combining both convolutional and transformer blocks. We revisit the key design principles of hybrid architectures and propose a simple and effective \emph{asymmetric} architecture, where the distribution of convolutional and transformer blocks is \emph{asymmetric}, containing more convolutional blocks in the earlier stages, followed by more transformer blocks in later stages. AsCAN supports a variety of tasks: recognition, segmentation, class-conditional image generation, and features a superior trade-off between performance and latency. We then scale the same architecture to solve a large-scale text-to-image task and show state-of-the-art performance compared to the most recent public and commercial models. Notably, even without any computation optimization for transformer blocks, our models still yield faster inference speed than existing works featuring efficient attention mechanisms, highlighting the advantages and the value of our approach.

Related papers

• Scalable, Tokenization-Free Diffusion Model Architectures with Efficient Initial Convolution and Fixed-Size Reusable Structures for On-Device Image Generation [0.0]
  Vision Transformers and U-Net architectures have been widely adopted in the implementation of Diffusion Models. We propose an architecture that utilizes a fixed-size, reusable transformer block as a core structure. Our architecture is characterized by low complexity, token-free design, absence of positional embeddings, uniformity, and scalability.
  arXiv  Detail & Related papers  (2024-11-09T08:58:57Z)
• Real-Time Image Segmentation via Hybrid Convolutional-Transformer Architecture Search [49.81353382211113]
  We address the challenge of integrating multi-head self-attention into high resolution representation CNNs efficiently. We develop a multi-target multi-branch supernet method, which fully utilizes the advantages of high-resolution features. We present a series of model via Hybrid Convolutional-Transformer Architecture Search (HyCTAS) method that searched for the best hybrid combination of light-weight convolution layers and memory-efficient self-attention layers.
  arXiv  Detail & Related papers  (2024-03-15T15:47:54Z)
• Perceiving Longer Sequences With Bi-Directional Cross-Attention Transformers [13.480259378415505]
  BiXT scales linearly with input size in terms of computational cost and memory consumption. BiXT is inspired by the Perceiver architectures but replaces iterative attention with an efficient bi-directional cross-attention module. By combining efficiency with the generality and performance of a full Transformer architecture, BiXT can process longer sequences.
  arXiv  Detail & Related papers  (2024-02-19T13:38:15Z)
• Energy-efficient Task Adaptation for NLP Edge Inference Leveraging Heterogeneous Memory Architectures [68.91874045918112]
  adapter-ALBERT is an efficient model optimization for maximal data reuse across different tasks. We demonstrate the advantage of mapping the model to a heterogeneous on-chip memory architecture by performing simulations on a validated NLP edge accelerator.
  arXiv  Detail & Related papers  (2023-03-25T14:40:59Z)
• Rich CNN-Transformer Feature Aggregation Networks for Super-Resolution [50.10987776141901]
  Recent vision transformers along with self-attention have achieved promising results on various computer vision tasks. We introduce an effective hybrid architecture for super-resolution (SR) tasks, which leverages local features from CNNs and long-range dependencies captured by transformers. Our proposed method achieves state-of-the-art SR results on numerous benchmark datasets.
  arXiv  Detail & Related papers  (2022-03-15T06:52:25Z)
• Multi-Exit Vision Transformer for Dynamic Inference [88.17413955380262]
  We propose seven different architectures for early exit branches that can be used for dynamic inference in Vision Transformer backbones. We show that each one of our proposed architectures could prove useful in the trade-off between accuracy and speed.
  arXiv  Detail & Related papers  (2021-06-29T09:01:13Z)
• Operation Embeddings for Neural Architecture Search [15.033712726016255]
  We propose the replacement of fixed operator encoding with learnable representations in the optimization process. Our method produces top-performing architectures that share similar operation and graph patterns.
  arXiv  Detail & Related papers  (2021-05-11T09:17:10Z)
• Towards Accurate and Compact Architectures via Neural Architecture Transformer [95.4514639013144]
  It is necessary to optimize the operations inside an architecture to improve the performance without introducing extra computational cost. We have proposed a Neural Architecture Transformer (NAT) method which casts the optimization problem into a Markov Decision Process (MDP) We propose a Neural Architecture Transformer++ (NAT++) method which further enlarges the set of candidate transitions to improve the performance of architecture optimization.
  arXiv  Detail & Related papers  (2021-02-20T09:38:10Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.