search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.59k stars 7.95k forks source link

My Average loss is not decreasing in yolov4-csp? #7674

Open akashAD98 opened 3 years ago

akashAD98 commented 3 years ago

I have 29 classes, so my max_batches=(29*2000=58,0000) so I trained my model for 58000 steps & my avg_loss is 24. So my question is after increasing the max_batches & training steps?? Will my loss decrease after few steps ??? the learning rate & max_delta will affect the model loss?? i selected max_delta=5 in last [yolo] layer & my lr_rate is 0.0001 please anyone can clarify my doubt. Thanks in advance.

poornimajd commented 3 years ago

Depends on how challenging is your dataset and also number of images you have.

akashAD98 commented 3 years ago

@poornimajd I think you are also working on Deep learning. can you please connect with me on https://www.linkedin.com/in/akash-desai-1b482b196 so we will discuss regarding this?

PedroBarcha commented 2 years ago

I have 29 classes, so my max_batches=(29*2000=58,0000) so I trained my model for 58000 steps & my avg_loss is 24. So my question is after increasing the max_batches & training steps?? Will my loss decrease after few steps ??? the learning rate & max_delta will affect the model loss?? i selected max_delta=5 in last [yolo] layer & my lr_rate is 0.0001 please anyone can clarify my doubt. Thanks in advance.

Same here, only when training with CSP. Found out why?

index

vokhidovhusan commented 2 years ago

I have 29 classes, so my max_batches=(29*2000=58,0000) so I trained my model for 58000 steps & my avg_loss is 24. So my question is after increasing the max_batches & training steps?? Will my loss decrease after few steps ??? the learning rate & max_delta will affect the model loss?? i selected max_delta=5 in last [yolo] layer & my lr_rate is 0.0001 please anyone can clarify my doubt. Thanks in advance.

I am facing the same issue. My avg loss also around 30 after 5k-10k iteration