marufahmed/dla_paper_resources

Supporting Resources for the paper 'Optimum Deep Learning method for Document Layout Analysis in Low Resource Languages' # Process to calculate Intersection Over Union of Deep Learning Based Document Layout Analysis for the paper 'Optimum Deep Learning method for Document Layout Analysis in Low Resource Languages' Resources ## Inference - Document Image Transformer. - Layout LM. - Custom Trained YoloV8 was inferred on local system ## Resize To normalize the pixel values of images, inferred images were resized using resize.ipynb. ## Annotation To create the annotations for inferred predictions, labelme was used. The classes that were labeled are as follows: - Paragraph - Image - Table After annotating the inferred images, corresponding json files were created. ## IOU comparison Individual & collective IOU comparison was measured using iou_comparison.ipynb ## Collected Data | | | bengali 1 | | bengali 2 | | marathi | | gujrati | | |---------------------------|-----------|----------|---------|----------|---------|---------|----------|---------|----------| | Model | Class | Overlapping Areas | Average IoU | Overlapping Areas | Average IoU | Overlapping Areas | Average IoU | Overlapping Areas | Average IoU | | Ground Truth | Paragraph | 14 | 1 | 6 | 1 | 10 | 1 | 17 | 1 | | Ground Truth | Table | 1 | 1 | 0 | 0 | 1 | 1 | 0 | 0 | | Ground Truth | Image | 0 | 0 | 1 | 1 | 0 | 0 | 2 | 1 | | Yolo V8 custom trained | Paragraph | 14 | 0.812 | 5 | 0.808 | 10 | 0.904 | 16 | 0.865 | | Yolo V8 custom trained | Table | 1 | 0.788 | 0 | 0 | 1 | 0.906 | 0 | 0 | | Yolo V8 custom trained | Image | 0 | 0 | 1 | 0.938 | 0 | 0 | 2 | 0.777 | | Document Image Transformer | Paragraph | 12 | 0.655 | 5 | 0.732 | 1 …