search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.77k stars 7.96k forks source link

YOLO for beginners #3148

Open maikatouzet opened 5 years ago

maikatouzet commented 5 years ago

Hello, I am a beginner and I am discovering YOLO. Therefore, I have some questions :

1- What is the main difference between YOLOv2 and YOLOv3 ? Are there some cases when YOLOv2 is more efficient than YOLOv3 ?

2- How does YOLO take the background into account ? How would it react if I decided to put an image with just the background (no label) in the dataset ? What if there are several objects on an image and I only label one of them ?

3- How is the mAP calculated ? And how can I have the mAP chart so I know which .weights file I should use ? (I saw several posts about compiling darknet with OPENCV=1 but it still doesn't seem to work)

4- Is there any other difference than the number of convolutional layers between YOLO and tiny YOLO ? As I would like to have instantaneous detection, I only use tiny YOLO which is faster, but is there a significant efficiency difference with basic YOLO ? Could I use YOLO on a dataset pretrained by tiny YOLO ?

5- Is it possible to associate numbers to the classes of the training so we don’t mix up between the different trainings ?

6- Can we associate YOLO and Caffee ? (By the way, if someone has some knowledge about Caffee, I'd be glad to have some explanations on what it is)

7- What happen if we stop a training and then want to restart it? Does it start again all over from the beginning? And what if we add images to a dataset we already trained? Is it possible to keep the first training and just complete it instead of restarting it al?

8- Can we convert a YOLOv3 training in YOLOv2?

Thank you!

Vic-TheGreat commented 5 years ago

lol.... first of all, this seems like a school homework and I am not sure you will find anyone willing to sit down and spell it out for you. You have to sit in front of a computer, open google and start to seek answers.

That been said, I propose that you ask these questions in different threads, as that will allow different experts of the different sub-question, see them and be able to assist you or point you in the right direction.

Bonne courage!!!

xzessmedia commented 5 years ago

im wondering why so many people use the git issue trackers as a forum/board.. its so annoying that most people are so lazy and stupid at the same time..

MASTER YOUR TOOLS