Skip to content

Latest commit

 

History

History
23 lines (11 loc) · 1.22 KB

weekly_report_0913_rcnn.md

File metadata and controls

23 lines (11 loc) · 1.22 KB

Xiaolong - Week of 09/13/2019

1. Papers and code

1.1 Papers Read

Rich feature hierarchies for accurate object detection and semantic segmentation: This paper introduced the R-CNN, an algorithm for object detection. R-CNN first generates some region proposals then apply CNN to classify each region.

1.2 Code Written

phspaceMetric.ipynb: I am trying to implement the metric to evaluate the observations in parameter space, defined by color, magnitude, and time gaps between two visits.

2. Figures

r-cnn

Figure 1: R-CNN object detection system: (1). takes an input image, (2). extracts around 2k region proposals, (3). computes features for each region using CNN and (4). classifies each region.

3. Results

I learned an object detection method, R-CNN, which first apply the selective search algorithm to extract region proposals and next use CNN to classify objects in each region. The problem with R-CNN is that it takes huge amount of time to train. This drawback is solved by Fast R-CNN and Faster R-CNN, which make possible the real time detection. R-CNN may be used to detect light echoes regardless their sizes inside a field.