EP-MAE: A resource-efficient masked autoencoding framework for 3D neural representation learning.

Scalable and Seamless: Accelerated Virtual-Tiling for High-Fidelity Hybrid CNN–Transformer Codecs

We propose a high-capacity, end-to-end framework for large-scale image compression that addresses the trade-off between tiling scalability and perceptual quality, a challenge stemming from the patch-based processing required for high-resolution inputs, which often introduces disruptive stitching artifacts. To mitigate this issue, we present a unified framework built on three complementary components: 1) Accelerated Virtual-Tiling, which simulates boundary interactions during training to improve spatial consistency without incurring the memory cost of multi-patch encoding; 2) Seam-Targeted Distance-Masked Self-Attention, a latent bottleneck mechanism that enables information exchange across patch boundaries; and 3) Boundary-Aware Regularization, which enforces consistency at tile interfaces through an explicit loss formulation. By explicitly modeling cross-boundary dependencies, the proposed method effectively suppresses stitching artifacts while maintaining scalability to high-resolution inputs. Extensive experiments on the Kodak, JPEG AI, and CLIC 2025 datasets demonstrate competitive or superior rate-distortion performance, achieving high structural fidelity with MS-SSIM values of approximately 0.998 at high compression ratios. These results indicate that the proposed framework provides an effective solution with substantially reduced boundary discontinuities for advanced neural image compression systems based on hybrid CNN-Transformer architectures.

AutoEncoder-Compressed Parallel Split Learning for Pre-trained Model Fine-Tuning

AE-PSL is proposed, a communication-efficient PSL framework that compresses intermediate activations and gradients using a lightweight AutoEncoder placed at the split layer and introduces a novel two-stage alignment mechanism, which adapts the AE to the pre-trained model's feature manifold and client-specific feature distributions before DFT.

Mage-Flow: An Efficient Native-Resolution Foundation Model for Image Generation and Editing

The results show that careful tokenizer--backbone--system co-design can deliver strong high-resolution generation and editing within an efficient 4B model family.

UvA-DARE (Digital Academic Repository) Elastic ViTs from Pretrained Models without Retraining

RADIANCE: Relative Adaptive Denoising with IP-Adapter for Novel Concept Enhancement

Text-to-image (T2I) diffusion models have achieved striking progress but still struggle to synthesize rare concepts involving unusual attribute-object pairings, often resulting in concept omission or semantic drift where a dominant entity overwhelms the generation. Tracing these failures to a lack of compositional balance during the denoising trajectory, we propose RADIANCE, a training-free framework that treats inference as a closed-loop feedback process. RADIANCE augments pretrained backbones with three modular components: (1) a Compositional Similarity Monitor (CSM) that tracks the emergence of objects and attributes in intermediate latents via CLIP-based feedback; (2) a Bidirectional Scale Controller (BSC) that applies a reactive"restoring force"using positive and negative IP-Adapter scales to rebalance biased trajectories; and (3) a Feedback Guidance Scheduler (FGS) that coordinates these updates across timesteps without additional training. We further extend the framework to multi-object prompts via Delayed Adapter Activation (DAA) and Layer-wise Alternating Guidance (LAG) to prevent premature concept fusion. By overlapping monitoring and denoising through pipelined execution, RADIANCE maintains competitive latency while significantly enhancing the per-sample success rate and effective throughput. Experiments on RareBench and T2I-CompBench demonstrate that RADIANCE consistently enhances compositional alignment and perceptual quality over state-of-the-art baselines.

VQ-Transplant: Efficient VQ-Module Integration for Pre-trained Visual Tokenizers

Vector Quantization (VQ) underpins modern discrete visual tokenization. However, training quantization modules for state-of-the-art VQ-based models requires significant computational resources which, in practice, all but prevents the development of novel, cutting-edge VQ techniques under resource constraints. To address this limitation, we propose {\bf VQ-Transplant}, a simple framework that enables plug-and-play integration of new VQ modules into frozen, pre-trained tokenizers by replacing their native VQ modules. Crucially, the proposed transplantation process preserves all encoder-decoder parameters, obviating the need for costly end-to-end retraining when modifying the quantization method. To mitigate decoder-quantization mismatch, we introduce a lightweight decoder adaptation strategy (trained for only 5 epochs on ImageNet-1k) to align feature priors with the new quantization space. In our empirical evaluation, we find that VQ-Transplant allows obtaining near state-of-the-art reconstruction fidelity for industry-level models like VAR while reducing the training cost by 95\%. VQ-Transplant democratizes quantization research by enabling resource-efficient integration of novel VQ techniques while matching industry-level reconstruction performance.