Focused on computer vision, image processing, color science, and AI algorithms; sharing hardcore tech, engineering practice, and deep insights as a diligent AI technology practitioner.
YOLO26 eliminates NMS and DFL, adopts a dual‑head design, MuSGD optimizer, progressive loss weighting, and STAL small‑object assignment, achieving 57.5 mAP with 1.7 ms latency on COCO while unifying detection, segmentation, pose, OBB and open‑set tasks, as shown by extensive ablations.
AnyFlow introduces a flow‑map distillation framework that enables video diffusion models to produce high‑quality results in just four steps while continuously improving with additional sampling steps, supporting both causal and bidirectional architectures up to 14 B parameters and allowing downstream fine‑tuning.
AnyFlow introduces a flow‑map distillation framework that enables video diffusion models to produce high‑quality results in just four sampling steps while still gaining quality as the number of steps increases, supporting both causal and bidirectional architectures and scaling up to 14 B parameters.
EUPE introduces a three‑stage “scale‑then‑shrink” distillation pipeline that first trains a large proxy model to absorb heterogeneous expert knowledge and then compresses it into an 86M encoder, achieving state‑of‑the‑art performance on image classification, dense prediction and vision‑language tasks on an iPhone with only 62 ms latency.
TCL Huaxing’s XRGB technology adds an independent cyan sub‑pixel to the traditional RGB layout, forming an RGBC four‑color architecture, custom cyan color‑filter and backlight, and a dedicated color‑mapping algorithm that together deliver BT.2020 131% gamut, 7000:1 contrast, 0.7% reflectivity and true‑4K resolution, redefining LCD display limits.
A large-scale benchmark of 20 pretrained ViT teachers across 11 families shows that attention copy and distillation improve some models but hurt others—especially DINOv2, CLIP, and BEiTv2—due to architecture mismatches, and adding the teachers' native components to students restores the lost performance.
The article analyzes ByteDance's Cola DLM, a fully open‑source continuous diffusion language model that abandons token‑centric generation in favor of latent semantic representations, detailing its architecture, training strategy, scaling stability, and how it compares with the earlier ELF model.
The paper presents a pixel‑level statistical study of the ASC StEM2 test film, building a three‑layer physical‑perceptual comparison of EXR, SDR and HDR masters, revealing that about 82 % of image regions can be restored through a restrained restoration process while the remaining areas require targeted semantic adjustments, offering concrete guidance for AI‑driven HDR conversion and industry standards.
Pixel‑Searcher introduces an agentic search‑driven visual perception framework that integrates web‑based evidence with pixel‑level grounding, and the new WebEyes benchmark demonstrates its superiority over existing open‑ and closed‑source multimodal models across localization, segmentation, and VQA tasks.
Qwen3-VL-Seg, the latest open‑source multimodal LLM from Alibaba, extends bounding‑box predictions to pixel‑accurate masks using a lightweight box‑guided decoder, achieving strong performance on both closed‑set and open‑world segmentation tasks with only 0.4% extra parameters.
