weekly_report_0927_yolo.md

Xiaolong - Week of 09/27/2019

1. Papers and code

1.1 Papers Read

Object Detection Algorithms:

• CNN: sliding window

• RCNN: regional based

• Fast RCNN:

• Faster RCNN:

• YOLO: This paper introduced the YOLO, an algorithm for real time object detection.

• RetinaNet:

1.2 Code Written

BloodCellDetection.ipynb: I run an example of application of YOLO on cola, it trains again the last layer of a pretrained yolo model on about 400 labeled images of red blood cells.

2. Figures

Figure 1: The YOLO Model. It divides the image into an $S\times S$ grid and for each grid cell pridicts $B$ bounding boxes ($b_x, b_y, b_h, b_w$), confidence for those boxes ($p_b$), and $C$ class probabilities. It runs an input through a CNN and output an $S\times S \times (B*5 + C)$ tensor.

Figure 2: The predicted bounding box for a picture contains many red blood cells.

3. Results

I learned the YOLO object detection algorithm. It runs an image through CNN and outputs a $S\times S \times (B*5 + C)$ tensor that contains the bounding box and confidence. Then it filters through all the boxes and selects the only one with highest probability.