TasksSotADatasetsPapersMethodsSubmitAbout
Papers With Code 2

A community resource for machine learning research: papers, code, benchmarks, and state-of-the-art results.

Explore

Notable BenchmarksAll SotADatasetsPapersMethods

Community

Submit ResultsAbout

Data sourced from the PWC Archive (CC-BY-SA 4.0). Built by the community, for the community.

Papers/Constrained R-CNN: A general image manipulation detection ...

Constrained R-CNN: A general image manipulation detection model

Chao Yang, Huizhou Li, Fangting Lin, Bin Jiang, Hao Zhao

2019-11-19Region ProposalImage ForensicsImage Manipulation LocalizationGeneral ClassificationImage ManipulationImage Manipulation Detection
PaperPDF

Abstract

Recently, deep learning-based models have exhibited remarkable performance for image manipulation detection. However, most of them suffer from poor universality of handcrafted or predetermined features. Meanwhile, they only focus on manipulation localization and overlook manipulation classification. To address these issues, we propose a coarse-to-fine architecture named Constrained R-CNN for complete and accurate image forensics. First, the learnable manipulation feature extractor learns a unified feature representation directly from data. Second, the attention region proposal network effectively discriminates manipulated regions for the next manipulation classification and coarse localization. Then, the skip structure fuses low-level and high-level information to refine the global manipulation features. Finally, the coarse localization information guides the model to further learn the finer local features and segment out the tampered region. Experimental results show that our model achieves state-of-the-art performance. Especially, the F1 score is increased by 28.4%, 73.2%, 13.3% on the NIST16, COVERAGE, and Columbia dataset.

Results

TaskDatasetMetricValueModel
Image Manipulation DetectionCOVERAGEAUC0.553CR-CNN
Image Manipulation DetectionCOVERAGEBalanced Accuracy0.391CR-CNN
Image Manipulation DetectionColumbiaAUC0.755CR-CNN
Image Manipulation DetectionColumbiaBalanced Accuracy0.631CR-CNN
Image Manipulation DetectionCocoGlideAUC0.589CR-CNN
Image Manipulation DetectionCocoGlideBalanced Accuracy0.447CR-CNN
Image Manipulation DetectionDSO-1AUC0.576CR-CNN
Image Manipulation DetectionDSO-1Balanced Accuracy0.289CR-CNN
Image Manipulation DetectionCasia V1+AUC0.67CR-CNN
Image Manipulation DetectionCasia V1+Balanced Accuracy0.481CR-CNN
VideoCOVERAGEAUC0.553CR-CNN
VideoCOVERAGEBalanced Accuracy0.391CR-CNN
VideoColumbiaAUC0.755CR-CNN
VideoColumbiaBalanced Accuracy0.631CR-CNN
VideoCocoGlideAUC0.589CR-CNN
VideoCocoGlideBalanced Accuracy0.447CR-CNN
VideoDSO-1AUC0.576CR-CNN
VideoDSO-1Balanced Accuracy0.289CR-CNN
VideoCasia V1+AUC0.67CR-CNN
VideoCasia V1+Balanced Accuracy0.481CR-CNN
Temporal Action LocalizationCOVERAGEAUC0.553CR-CNN
Temporal Action LocalizationCOVERAGEBalanced Accuracy0.391CR-CNN
Temporal Action LocalizationColumbiaAUC0.755CR-CNN
Temporal Action LocalizationColumbiaBalanced Accuracy0.631CR-CNN
Temporal Action LocalizationCocoGlideAUC0.589CR-CNN
Temporal Action LocalizationCocoGlideBalanced Accuracy0.447CR-CNN
Temporal Action LocalizationDSO-1AUC0.576CR-CNN
Temporal Action LocalizationDSO-1Balanced Accuracy0.289CR-CNN
Temporal Action LocalizationCasia V1+AUC0.67CR-CNN
Temporal Action LocalizationCasia V1+Balanced Accuracy0.481CR-CNN
Anomaly DetectionCOVERAGEAUC0.553CR-CNN
Anomaly DetectionCOVERAGEBalanced Accuracy0.391CR-CNN
Anomaly DetectionColumbiaAUC0.755CR-CNN
Anomaly DetectionColumbiaBalanced Accuracy0.631CR-CNN
Anomaly DetectionCocoGlideAUC0.589CR-CNN
Anomaly DetectionCocoGlideBalanced Accuracy0.447CR-CNN
Anomaly DetectionDSO-1AUC0.576CR-CNN
Anomaly DetectionDSO-1Balanced Accuracy0.289CR-CNN
Anomaly DetectionCasia V1+AUC0.67CR-CNN
Anomaly DetectionCasia V1+Balanced Accuracy0.481CR-CNN
Zero-Shot LearningCOVERAGEAUC0.553CR-CNN
Zero-Shot LearningCOVERAGEBalanced Accuracy0.391CR-CNN
Zero-Shot LearningColumbiaAUC0.755CR-CNN
Zero-Shot LearningColumbiaBalanced Accuracy0.631CR-CNN
Zero-Shot LearningCocoGlideAUC0.589CR-CNN
Zero-Shot LearningCocoGlideBalanced Accuracy0.447CR-CNN
Zero-Shot LearningDSO-1AUC0.576CR-CNN
Zero-Shot LearningDSO-1Balanced Accuracy0.289CR-CNN
Zero-Shot LearningCasia V1+AUC0.67CR-CNN
Zero-Shot LearningCasia V1+Balanced Accuracy0.481CR-CNN
Activity RecognitionCOVERAGEAUC0.553CR-CNN
Activity RecognitionCOVERAGEBalanced Accuracy0.391CR-CNN
Activity RecognitionColumbiaAUC0.755CR-CNN
Activity RecognitionColumbiaBalanced Accuracy0.631CR-CNN
Activity RecognitionCocoGlideAUC0.589CR-CNN
Activity RecognitionCocoGlideBalanced Accuracy0.447CR-CNN
Activity RecognitionDSO-1AUC0.576CR-CNN
Activity RecognitionDSO-1Balanced Accuracy0.289CR-CNN
Activity RecognitionCasia V1+AUC0.67CR-CNN
Activity RecognitionCasia V1+Balanced Accuracy0.481CR-CNN
Action LocalizationCOVERAGEAUC0.553CR-CNN
Action LocalizationCOVERAGEBalanced Accuracy0.391CR-CNN
Action LocalizationColumbiaAUC0.755CR-CNN
Action LocalizationColumbiaBalanced Accuracy0.631CR-CNN
Action LocalizationCocoGlideAUC0.589CR-CNN
Action LocalizationCocoGlideBalanced Accuracy0.447CR-CNN
Action LocalizationDSO-1AUC0.576CR-CNN
Action LocalizationDSO-1Balanced Accuracy0.289CR-CNN
Action LocalizationCasia V1+AUC0.67CR-CNN
Action LocalizationCasia V1+Balanced Accuracy0.481CR-CNN
3D Action RecognitionCOVERAGEAUC0.553CR-CNN
3D Action RecognitionCOVERAGEBalanced Accuracy0.391CR-CNN
3D Action RecognitionColumbiaAUC0.755CR-CNN
3D Action RecognitionColumbiaBalanced Accuracy0.631CR-CNN
3D Action RecognitionCocoGlideAUC0.589CR-CNN
3D Action RecognitionCocoGlideBalanced Accuracy0.447CR-CNN
3D Action RecognitionDSO-1AUC0.576CR-CNN
3D Action RecognitionDSO-1Balanced Accuracy0.289CR-CNN
3D Action RecognitionCasia V1+AUC0.67CR-CNN
3D Action RecognitionCasia V1+Balanced Accuracy0.481CR-CNN
Action RecognitionCOVERAGEAUC0.553CR-CNN
Action RecognitionCOVERAGEBalanced Accuracy0.391CR-CNN
Action RecognitionColumbiaAUC0.755CR-CNN
Action RecognitionColumbiaBalanced Accuracy0.631CR-CNN
Action RecognitionCocoGlideAUC0.589CR-CNN
Action RecognitionCocoGlideBalanced Accuracy0.447CR-CNN
Action RecognitionDSO-1AUC0.576CR-CNN
Action RecognitionDSO-1Balanced Accuracy0.289CR-CNN
Action RecognitionCasia V1+AUC0.67CR-CNN
Action RecognitionCasia V1+Balanced Accuracy0.481CR-CNN
Image Manipulation LocalizationColumbiaAverage Pixel F1(Fixed threshold)0.631CR-CNN
Image Manipulation LocalizationCOVERAGEAverage Pixel F1(Fixed threshold)0.391CR-CNN
Image Manipulation LocalizationCasia V1+Average Pixel F1(Fixed threshold)0.481CR-CNN
Image Manipulation LocalizationCocoGlideAverage Pixel F1(Fixed threshold)0.447CR-CNN
Image Manipulation LocalizationDSO-1Average Pixel F1(Fixed threshold)0.289CR-CNN

Related Papers

Beyond Fully Supervised Pixel Annotations: Scribble-Driven Weakly-Supervised Framework for Image Manipulation Localization2025-07-17Bridging Annotation Gaps: Transferring Labels to Align Object Detection Datasets2025-06-05Towards Reliable Identification of Diffusion-based Image Manipulations2025-06-05UniWorld-V1: High-Resolution Semantic Encoders for Unified Visual Understanding and Generation2025-06-03Weakly-supervised Localization of Manipulated Image Regions Using Multi-resolution Learned Features2025-05-29Swin Transformer for Robust CGI Images Detection: Intra- and Inter-Dataset Analysis across Multiple Color Spaces2025-05-22Can ChatGPT Perform Image Splicing Detection? A Preliminary Study2025-05-22RBench-V: A Primary Assessment for Visual Reasoning Models with Multi-modal Outputs2025-05-22