DiC: Rethinking Conv3x3 Designs in Diffusion Models

• URL: http://arxiv.org/abs/2501.00603v1
• Date: Tue, 31 Dec 2024 19:00:01 GMT
• Title: DiC: Rethinking Conv3x3 Designs in Diffusion Models
• Authors: Yuchuan Tian, Jing Han, Chengcheng Wang, Yuchen Liang, Chao Xu, Hanting Chen,
• Abstract summary: Diffusion models have shown exceptional performance in visual generation tasks. Recent models have shifted from traditional U-Shaped CNN-Attention hybrid structures to fully transformer-based isotropic architectures. We construct a scaled-up purely convolutional diffusion model using 3x3 Convolution.
• Score: 19.902725825796836
• License:
• Abstract: Diffusion models have shown exceptional performance in visual generation tasks. Recently, these models have shifted from traditional U-Shaped CNN-Attention hybrid structures to fully transformer-based isotropic architectures. While these transformers exhibit strong scalability and performance, their reliance on complicated self-attention operation results in slow inference speeds. Contrary to these works, we rethink one of the simplest yet fastest module in deep learning, 3x3 Convolution, to construct a scaled-up purely convolutional diffusion model. We first discover that an Encoder-Decoder Hourglass design outperforms scalable isotropic architectures for Conv3x3, but still under-performing our expectation. Further improving the architecture, we introduce sparse skip connections to reduce redundancy and improve scalability. Based on the architecture, we introduce conditioning improvements including stage-specific embeddings, mid-block condition injection, and conditional gating. These improvements lead to our proposed Diffusion CNN (DiC), which serves as a swift yet competitive diffusion architecture baseline. Experiments on various scales and settings show that DiC surpasses existing diffusion transformers by considerable margins in terms of performance while keeping a good speed advantage. Project page: https://github.com/YuchuanTian/DiC

Related papers

• Atleus: Accelerating Transformers on the Edge Enabled by 3D Heterogeneous Manycore Architectures [18.355570259898]
  We propose the design of a 3D heterogeneous architecture referred to as Atleus. Atleus incorporates heterogeneous computing resources specifically optimized to accelerate transformer models. We show that Atleus outperforms existing state-of-the-art by up to 56x and 64.5x in terms of performance and energy efficiency respectively.
  arXiv  Detail & Related papers  (2025-01-16T15:11:33Z)
• TinyFusion: Diffusion Transformers Learned Shallow [52.96232442322824]
  Diffusion Transformers have demonstrated remarkable capabilities in image generation but often come with excessive parameterization. We present TinyFusion, a depth pruning method designed to remove redundant layers from diffusion transformers via end-to-end learning. Experiments with DiT-XL show that TinyFusion can craft a shallow diffusion transformer at less than 7% of the pre-training cost, achieving a 2$times$ speedup with an FID score of 2.86.
  arXiv  Detail & Related papers  (2024-12-02T07:05:39Z)
• Kolmogorov-Arnold Transformer [72.88137795439407]
  We introduce the Kolmogorov-Arnold Transformer (KAT), a novel architecture that replaces layers with Kolmogorov-Arnold Network (KAN) layers. We identify three key challenges: (C1) Base function, (C2) Inefficiency, and (C3) Weight. With these designs, KAT outperforms traditional-based transformers.
  arXiv  Detail & Related papers  (2024-09-16T17:54:51Z)
• ToddlerDiffusion: Interactive Structured Image Generation with Cascaded Schrödinger Bridge [63.00793292863]
  ToddlerDiffusion is a novel approach to decomposing the complex task of RGB image generation into simpler, interpretable stages. Our method, termed ToddlerDiffusion, cascades modality-specific models, each responsible for generating an intermediate representation. ToddlerDiffusion consistently outperforms state-of-the-art methods.
  arXiv  Detail & Related papers  (2023-11-24T15:20:01Z)
• FastViT: A Fast Hybrid Vision Transformer using Structural Reparameterization [14.707312504365376]
  We introduce FastViT, a hybrid vision transformer architecture that obtains the state-of-the-art latency-accuracy trade-off. We show that our model is 3.5x faster than CMT, 4.9x faster than EfficientNet, and 1.9x faster than ConvNeXt on a mobile device for the same accuracy on the ImageNet dataset.
  arXiv  Detail & Related papers  (2023-03-24T17:58:32Z)
• Efficient Neural Net Approaches in Metal Casting Defect Detection [0.0]
  This research proposes a lightweight architecture that is efficient in terms of accuracy and inference time. Our results indicate that a custom model of 590K parameters with depth-wise separable convolutions outperformed pretrained architectures.
  arXiv  Detail & Related papers  (2022-08-08T13:54:36Z)
• Squeezeformer: An Efficient Transformer for Automatic Speech Recognition [99.349598600887]
  Conformer is the de facto backbone model for various downstream speech tasks based on its hybrid attention-convolution architecture. We propose the Squeezeformer model, which consistently outperforms the state-of-the-art ASR models under the same training schemes.
  arXiv  Detail & Related papers  (2022-06-02T06:06:29Z)
• SideRT: A Real-time Pure Transformer Architecture for Single Image Depth Estimation [11.513054537848227]
  We propose a pure transformer architecture called SideRT that can attain excellent predictions in real-time. This is the first work to show that transformer-based networks can attain state-of-the-art performance in real-time in the single image depth estimation field.
  arXiv  Detail & Related papers  (2022-04-29T05:46:20Z)
• TCCT: Tightly-Coupled Convolutional Transformer on Time Series Forecasting [6.393659160890665]
  We propose the concept of tightly-coupled convolutional Transformer(TCCT) and three TCCT architectures. Our experiments on real-world datasets show that our TCCT architectures could greatly improve the performance of existing state-of-art Transformer models.
  arXiv  Detail & Related papers  (2021-08-29T08:49:31Z)
• RT3D: Achieving Real-Time Execution of 3D Convolutional Neural Networks on Mobile Devices [57.877112704841366]
  This paper proposes RT3D, a model compression and mobile acceleration framework for 3D CNNs. For the first time, real-time execution of 3D CNNs is achieved on off-the-shelf mobiles.
  arXiv  Detail & Related papers  (2020-07-20T02:05:32Z)
• A Real-time Action Representation with Temporal Encoding and Deep Compression [115.3739774920845]
  We propose a new real-time convolutional architecture, called Temporal Convolutional 3D Network (T-C3D), for action representation. T-C3D learns video action representations in a hierarchical multi-granularity manner while obtaining a high process speed. Our method achieves clear improvements on UCF101 action recognition benchmark against state-of-the-art real-time methods by 5.4% in terms of accuracy and 2 times faster in terms of inference speed with a less than 5MB storage model.
  arXiv  Detail & Related papers  (2020-06-17T06:30:43Z)

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.