241121

1. YOLO-CLS: self made model

 - epochs = 1

 - conf. = 0.25

 - LoU = 0.7

 - Images = 1800

the model doesn't recognize me

 

2. YOLO: self made model

 - epochs = 1

 - conf. = 0.25

 - LoU = 0.7

 - Images = 1800

the model recognize everyone as me

 

3. Group project:

upgrading YOLO model by increasing the amount of dataset and controlling hyperparameter

 - the amount of photos of my face 1800 -> 4000

 - epochs from 1 to 30

 

1) case1

 - epochs = 16

 - conf. = 0.25

 - LoU = 0.7

 - total amount of images = 26700

 

 

2) case2

 - epochs = 30

 - conf. = 0.25

 - LoU = 0.7

 - total amount of images = 26700

 

 

 

4. Conclusion

 - we focused on getting various conditions of photos of my face,

   because members of our group realized that the various datasets really affect the model performance

 - and thanks to our effort, the model recognize my face as 'my_face' exactly

 - it seems like that the amount of the images for training a model really matters

 - even the model with epochs 30 sometimes recognize other people as me

 - hair affects the performance of a model

 

 - in a model with high epochs(30), the model distinguishes faces more accurately

 - in a model with low epochs(16), the model finds faces better

 - unfortunately, i don't get how the models work like this

 - it's just the learning from experience and our results are consistent with a theory

 

 - controlling hyperparameter affects model performance