Image Prediction Methods: Citation Approaches and Performance Comparison

This table provides an overview of different citation methods used in various approaches for image prediction. It includes the approach used, description, and prediction time per image for each method.\n\n| Citation Method | Approach Used | Description | Prediction Time/Image |\n|---|---|---|---|\n| 102 CNN (1988) | Image is divided into multiple regions; then each region is classified into various classes. | Lots of regions are required for accurate prediction, high computation time | — |\n| 18 R-CNN (2014) | Selective search algorithm | Selective search18 algorithm is used to generate regions, 2000 regions are extracted from each image | 40-50 s |\n| 103 SPP-Net (2014) | Spatial pyramid pooling layer | Eliminates repetitive processing of candidate region and fixed-length output will be generated | — |\n| 104 Fast R-CNN (2015) | ROI for generating feature vector and SoftMax classifier | Faster training and testing time with any input image size | 2 s |\n| 96 Faster R-CNN (2016) | RPN (region-based proposed network) | Replacement of selective search algorithm by RPN makes it faster algorithm. | 0.2 s |\n| 105 SSD | Multiscale bounding boxes | Small filters are applied to feature maps and predictions are made at different scales | 59 frames/s on standard datasets such as Pascal VOC and COCO |\n| 106 YOLO | Use single ConvNet | Classes and bounding box of the whole image will be predicted. | 45 frames/s |