Скачать с ютуб C 5.8 | Non Max Suppression | NMS | CNN | Object Detection | Machine learning | EvODN в хорошем качестве

C 5.8 | Non Max Suppression | NMS | CNN | Object Detection | Machine learning | EvODN

This is a repeat video of NMS, in case you have not taken the Classical CV (3rd) chapter:    • C38 | NMS | Non Max Suppression | Obj...   as you will see in this video, there will be multiple detections per object. So, how do we select the best out of those multiple detections. One option is to use confidence score thresholds. But we will see that, we cant have a single threshold for all objects. So, what we do is use Non Max Suppression. Basically keep the box with the highest confidence score and eliminate others that overlap this box with an IOU of more than, say, 50% or 70%. So, we use Sliding Windows to identify objects at different locations, Image Pyramid to identify objects of different sizes. These 2 techniques are applied at the input side. At the output side, we will have multiple detections per object and some invalid detections mostly on the background regions of the image with very low confidence scores. So, we first do a confidence score thresholding to eliminate detections on background regions of the image. Then we apply NMS to get the one best detection for each of the objects in the image. These are the 2 techniques used at the output side. ------------------------ This is a part of the course 'Evolution of Object Detection Networks'. See full playlist here:    • Evolution Of Object Detection Networks   ------------------------ Slide credit:    • Lecture 18- Deformable Part Models (D...   Dalal & Triggs Detector: https://lear.inrialpes.fr/people/trig... Image Credit - https://federerfan07.com/2011/06/25/l... Copyright Disclaimer: Under section 107 of the Copyright Act 1976, allowance is made for “fair use” for purposes such as criticism, comment, news reporting, teaching, scholarship, education and research.