zjx0101/ObjectClear

[CVPR'26] ObjectClear: Precise Object and Effect Removal with Adaptive Target-Aware Attention

What it solves

ObjectClear addresses the challenge of removing objects from images while also eliminating their associated effects (such as shadows or reflections) and maintaining the consistency of the background.

How it works

Built on SDXL-Inpainting and utilizing a CLIP image encoder to identify the target object, the model employs Adaptive Target-Aware Attention to jointly remove both the object and its effects. It is trained on the OBER (OBject-Effect Removal) dataset, a hybrid collection of camera-captured and simulated data. Users provide an image and a segmentation mask of the object to be removed, and the model generates a clean background.

Who it’s for

This tool is designed for researchers and developers working in image editing, computer vision, and generative AI, specifically those focused on object removal and inpainting.

Highlights

  • Joint Removal: Simultaneously eliminates the target object and its related visual effects.
  • Adaptive Target-Aware Attention: A specialized attention mechanism to improve removal precision.
  • OBER Dataset: Includes a newly released hybrid dataset specifically for object-effect removal research.
  • ReMOVE+ Metric: Introduces an improved evaluation metric to better assess consistency between the removed region and the surrounding background.

Related

  • Dispatch
    Hugging Face Object Detection Leaderboard

    Hugging Face released an Object Detection Leaderboard that ranks open-source models using COCO-style metrics and published a blog explaining how Average Precision and Average Recall are computed and what factors can influence the results.

  • Project
    RizwanMunawar/yolov7-object-tracking

    A computer vision project that integrates YOLOv7 and YOLOv8 for real-time object detection and tracking across video streams from various sources.

  • Project
    melandlabs/openloomi

    OpenLoomi is an open‑source, cross‑platform desktop “attention agent” that continuously gathers context from your work tools (email, chat, issue trackers, docs) into a searchable, encrypted memory. It surfaces reminders and lets you ask an LLM for help directly inside messaging apps. The runtime is agent‑agnostic, offering Skills and Plugins that integrate with Claude Code, Codex, OpenCode, Hermes, OpenClaw, etc. Installable via simple binaries, it emphasizes local‑first privacy and works out‑of‑the‑box.

  • Project
    roboflow/sports

    A collection of computer vision tools and datasets for sports analytics, focusing on challenges like ball tracking, player re-identification, and jersey number OCR.

  • Project
    koide3/direct_visual_lidar_calibration

    A target-less, automatic toolbox for LiDAR-camera extrinsic calibration that uses environment structure and texture for precise sensor alignment.