A counter-intuitive fact
Johnson Lindenstrauss Lemma
Proof
References
- course stuffs/Johnson-Lindenstrauss_Theorem)
Anatomy of YOLO Detector
Sliding window or attention mechanism ?
Idea of YOLO Detector
Classic object detection task normally has two steps, which are classify the content of region of interests (ROI) by e.g. CNN, and locate the ROI by e.g. “scanning with sliding window”.
Yolo uses a reasonable strong spatial constrains to merge these two steps into one step, and process the original image only once (input image to NN once), which literally means “You Only Look Once” (YOLO).
For the short explanation, Yolo divides the image into $S \times S$ grids, and suppose each cell of the grid take responsibility for the detection of objectness and classification of objects.
Bayes Inference by Hand
Too much information every day, I have to backup some sections of my brain here.
An interesting quiz
In a course by Prof. Isbell and Prof. Littman, I found a quiz as quite an good example to understand Bayes inference. Hence I post my deduction here as an archive.
The quiz is, there are two black boxes. In box 1 there are 3 green balls and 1 orange ball. In box 2 there are 3 blue balls and 2 green balls. The question is, when we randomly picked balls from one unknown box, we noticed that the first ball is green, what is the probability that the second picked ball is blue. Note, here $Pr(box=1)=Pr(box=2)=0.5$ is given.
