JanusDNA: A Powerful Bi-directional Hybrid DNA Foundation Model

Large language models (LLMs) have revolutionized natural language processing and are increasingly applied to other sequential data types, including genetic sequences. However, adapting LLMs to genomics presents significant challenges. Capturing complex genomic interactions requires modeling long-range dependencies within DNA sequences, where interactions often span over 10,000 base pairs, even within a single gene, posing substantial computational burdens under conventional model architectures and training paradigms. Moreover, standard LLM training approaches are suboptimal for DNA: autoregressive training, while efficient, supports only unidirectional understanding. However, DNA is inherently bidirectional, e.g., bidirectional promoters regulate transcription in both directions and account for nearly 11% of human gene expression. Masked language models (MLMs) allow bidirectional understanding but are inefficient, as only masked tokens contribute to the loss per step. To address these limitations, we introduce JanusDNA, the first bidirectional DNA foundation model built upon a novel pretraining paradigm that combines the optimization efficiency of autoregressive modeling with the bidirectional comprehension of masked modeling. JanusDNA adopts a hybrid Mamba, Attention and Mixture of Experts (MoE) architecture, combining long-range modeling of Attention with efficient sequential learning of Mamba. MoE layers further scale model capacity via sparse activation while keeping computational cost low. Notably, JanusDNA processes up to 1 million base pairs at single nucleotide resolution on a single 80GB GPU. Extensive experiments and ablations show JanusDNA achieves new SOTA results on three genomic representation benchmarks, outperforming models with 250x more activated parameters. Code: https://github.com/Qihao-Duan/JanusDNA

Difflare: Removing Image Lens Flare with Latent Diffusion Model

The recovery of high-quality images from images corrupted by lens flare presents a significant challenge in low-level vision. Contemporary deep learning methods frequently entail training a lens flare removing model from scratch. However, these methods, despite their noticeable success, fail to utilize the generative prior learned by pre-trained models, resulting in unsatisfactory performance in lens flare removal. Furthermore, there are only few works considering the physical priors relevant to flare removal. To address these issues, we introduce Difflare, a novel approach designed for lens flare removal. To leverage the generative prior learned by Pre-Trained Diffusion Models (PTDM), we introduce a trainable Structural Guidance Injection Module (SGIM) aimed at guiding the restoration process with PTDM. Towards more efficient training, we employ Difflare in the latent space. To address information loss resulting from latent compression and the stochastic sampling process of PTDM, we introduce an Adaptive Feature Fusion Module (AFFM), which incorporates the Luminance Gradient Prior (LGP) of lens flare to dynamically regulate feature extraction. Extensive experiments demonstrate that our proposed Difflare achieves state-of-the-art performance in real-world lens flare removal, restoring images corrupted by flare with improved fidelity and perceptual quality. The codes will be released soon.