Objectives
- Collect and preprocess a diverse dataset of forged RGB images with ground truth labels. - Perform data augmentation to increase dataset size and variety. - Implement YOLO v8n for detecting forged objects in RGB images using pre-trained COCO weights. - Train the model, optimizing hyperparameters for accuracy. - Evaluate model performance on a validation dataset and in challenging scenarios. - Develop a cross-platform mobile app with Flutter, integrating the trained model. - Enable image selection and forgery detection in the app with bounding boxes. - Test the app for reliability, performance, and user experience; address issues. - Deploy the app to app stores. - Compare the developed approach with existing forgery detection methods. - Analyze strengths, limitations, and areas for improvement.
Socio-Economic Benefit
- Can help in detection of fake objects - Can help in stoping fake images - Can help in identifying edited areas in Image - Anyone can use it easily
Methodologies
Dataset Collection Labeling Dataset export in YOLO Format Use YOLO v8n evaluate the model Develop app in Flutter Integrate the model in Flutter App
Outcome
Model Accuracy is 83% Works on almost all android devices Can Detect fake objects in RGB Images
Project Team Members
| Registration# | Name |
|---|---|
| CU-1220-2020 | Musawer Hussain |
| CU-1222-2020 | Izhar Badshah |
| CU-1230-2020 | Abdullah |
PROJECT GALLERY
DEEPFORGE
23 May, 2024 09:05 am
DEEPFORGE
23 May, 2024 09:05 am
DEEPFORGE
23 May, 2024 09:05 am
DEEPFORGE
23 May, 2024 09:05 am
DEEPFORGE
23 May, 2024 09:05 am
