Transformer is leading a trend in the field of image processing. Despite the great success that existing lightweight image processing transformers have achieved, they are tailored to FLOPs or parameters reduction, rather than practical inference acceleration. In this paper, we present a latency-aware image processing transformer, termed LIPT. We devise the low-latency proportion LIPT block that substitutes memory-intensive operators with the combination of self-attention and convolutions to achieve practical speedup. Specifically, we propose a novel non-volatile sparse masking self-attention (NVSM-SA) that utilizes a pre-computing sparse mask to capture contextual information from a larger window with no extra computation overload. Besides, a high-frequency reparameterization module (HRM) is proposed to make LIPT block reparameterization friendly, which improves the model's detail reconstruction capability. Extensive experiments on multiple image processing tasks (e.g., image super-resolution (SR), JPEG artifact reduction, and image denoising) demonstrate the superiority of LIPT on both latency and PSNR. LIPT achieves real-time GPU inference with state-of-the-art performance on multiple image SR benchmarks.
Knowledge distillation (KD) is a promising yet challenging model compression technique that transfers rich learning representations from a well-performing but cumbersome teacher model to a compact student model. Previous methods for image super-resolution (SR) mostly compare the feature maps directly or after standardizing the dimensions with basic algebraic operations (e.g. average, dot-product). However, the intrinsic semantic differences among feature maps are overlooked, which are caused by the disparate expressive capacity between the networks. This work presents MiPKD, a multi-granularity mixture of prior KD framework, to facilitate efficient SR model through the feature mixture in a unified latent space and stochastic network block mixture. Extensive experiments demonstrate the effectiveness of the proposed MiPKD method.
In image restoration (IR), leveraging semantic priors from segmentation models has been a common approach to improve performance. The recent segment anything model (SAM) has emerged as a powerful tool for extracting advanced semantic priors to enhance IR tasks. However, the computational cost of SAM is prohibitive for IR, compared to existing smaller IR models. The incorporation of SAM for extracting semantic priors considerably hampers the model inference efficiency. To address this issue, we propose a general framework to distill SAM's semantic knowledge to boost exiting IR models without interfering with their inference process. Specifically, our proposed framework consists of the semantic priors fusion (SPF) scheme and the semantic priors distillation (SPD) scheme. SPF fuses two kinds of information between the restored image predicted by the original IR model and the semantic mask predicted by SAM for the refined restored image. SPD leverages a self-distillation manner to distill the fused semantic priors to boost the performance of original IR models. Additionally, we design a semantic-guided relation (SGR) module for SPD, which ensures semantic feature representation space consistency to fully distill the priors. We demonstrate the effectiveness of our framework across multiple IR models and tasks, including deraining, deblurring, and denoising.
In an effort to reduce the computational load of Transformers, research on linear attention has gained significant momentum. However, the improvement strategies for attention mechanisms typically necessitate extensive retraining, which is impractical for large language models with a vast array of parameters. In this paper, we present DiJiang, a novel Frequency Domain Kernelization approach that enables the transformation of a pre-trained vanilla Transformer into a linear complexity model with little training costs. By employing a weighted Quasi-Monte Carlo method for sampling, the proposed approach theoretically offers superior approximation efficiency. To further reduce the training computational complexity, our kernelization is based on Discrete Cosine Transform (DCT) operations. Extensive experiments demonstrate that the proposed method achieves comparable performance to the original Transformer, but with significantly reduced training costs and much faster inference speeds. Our DiJiang-7B achieves comparable performance with LLaMA2-7B on various benchmark while requires only about 1/50 training cost. Code is available at https://github.com/YuchuanTian/DiJiang.
Recent advances have demonstrated the powerful capability of transformer architecture in image restoration. However, our analysis indicates that existing transformerbased methods can not establish both exact global and local dependencies simultaneously, which are much critical to restore the details and missing content of degraded images. To this end, we present an efficient image processing transformer architecture with hierarchical attentions, called IPTV2, adopting a focal context self-attention (FCSA) and a global grid self-attention (GGSA) to obtain adequate token interactions in local and global receptive fields. Specifically, FCSA applies the shifted window mechanism into the channel self-attention, helps capture the local context and mutual interaction across channels. And GGSA constructs long-range dependencies in the cross-window grid, aggregates global information in spatial dimension. Moreover, we introduce structural re-parameterization technique to feed-forward network to further improve the model capability. Extensive experiments demonstrate that our proposed IPT-V2 achieves state-of-the-art results on various image processing tasks, covering denoising, deblurring, deraining and obtains much better trade-off for performance and computational complexity than previous methods. Besides, we extend our method to image generation as latent diffusion backbone, and significantly outperforms DiTs.
Recent advancements in large language models have sparked interest in their extraordinary and near-superhuman capabilities, leading researchers to explore methods for evaluating and optimizing these abilities, which is called superalignment. In this context, our paper delves into the realm of vision foundation models, focusing on the concept of weak-to-strong generalization, which involves using a weaker model to supervise a stronger one, aiming to enhance the latter's capabilities beyond the former's limits. We introduce a novel and adaptively adjustable loss function for weak-to-strong supervision. Our comprehensive experiments span various scenarios, including few-shot learning, transfer learning, noisy label learning, and common knowledge distillation settings. The results are striking: our approach not only exceeds the performance benchmarks set by strong-to-strong generalization but also surpasses the outcomes of fine-tuning strong models with whole datasets. This compelling evidence underscores the significant potential of weak-to-strong generalization, showcasing its capability to substantially elevate the performance of vision foundation models. The code is available at https://github.com/ggjy/vision_weak_to_strong.
The recent trend of large language models (LLMs) is to increase the scale of both model size (\aka the number of parameters) and dataset to achieve better generative ability, which is definitely proved by a lot of work such as the famous GPT and Llama. However, large models often involve massive computational costs, and practical applications cannot afford such high prices. However, the method of constructing a strong model architecture for LLMs is rarely discussed. We first analyze the state-of-the-art language model architectures and observe the feature collapse problem. Based on the theoretical analysis, we propose that the nonlinearity is also very important for language models, which is usually studied in convolutional neural networks for vision tasks. The series informed activation function is then introduced with tiny calculations that can be ignored, and an augmented shortcut is further used to enhance the model nonlinearity. We then demonstrate that the proposed approach is significantly effective for enhancing the model nonlinearity through carefully designed ablations; thus, we present a new efficient model architecture for establishing modern, namely, PanGu-$\pi$. Experiments are then conducted using the same dataset and training strategy to compare PanGu-$\pi$ with state-of-the-art LLMs. The results show that PanGu-$\pi$-7B can achieve a comparable performance to that of benchmarks with about 10\% inference speed-up, and PanGu-$\pi$-1B can achieve state-of-the-art performance in terms of accuracy and efficiency. In addition, we have deployed PanGu-$\pi$-7B in the high-value domains of finance and law, developing an LLM named YunShan for practical application. The results show that YunShan can surpass other models with similar scales on benchmarks.
Post-training quantization (PTQ) has driven attention to producing efficient large language models (LLMs) with ultra-low costs. Since hand-craft quantization parameters lead to low performance in low-bit quantization, recent methods optimize the quantization parameters through block-wise reconstruction between the floating-point and quantized models. However, these methods suffer from two challenges: accumulated errors from independent one-by-one block quantization and reconstruction difficulties from extreme weight and activation outliers. To address these two challenges, we propose CBQ, a cross-block reconstruction-based PTQ method for LLMs. To reduce error accumulation, we introduce a cross-block dependency with the aid of a homologous reconstruction scheme to build the long-range dependency between adjacent multi-blocks with overlapping. To reduce reconstruction difficulty, we design a coarse-to-fine pre-processing (CFP) to truncate weight outliers and dynamically scale activation outliers before optimization, and an adaptive rounding scheme, called LoRA-Rounding, with two low-rank learnable matrixes to further rectify weight quantization errors. Extensive experiments demonstrate that: (1) CBQ pushes both activation and weight quantization to low-bit settings W4A4, W4A8, and W2A16. (2) CBQ achieves better performance than the existing state-of-the-art methods on various LLMs and benchmark datasets.